xref: /freebsd/sys/net/iflib.c (revision f7594707509b4d6169a6c7039ee640c0e45ef05a)
1 /*-
2  * Copyright (c) 2014-2018, Matthew Macy <mmacy@mattmacy.io>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions are met:
7  *
8  *  1. Redistributions of source code must retain the above copyright notice,
9  *     this list of conditions and the following disclaimer.
10  *
11  *  2. Neither the name of Matthew Macy nor the names of its
12  *     contributors may be used to endorse or promote products derived from
13  *     this software without specific prior written permission.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
16  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
19  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25  * POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 #include "opt_acpi.h"
34 #include "opt_sched.h"
35 
36 #include <sys/param.h>
37 #include <sys/types.h>
38 #include <sys/bus.h>
39 #include <sys/eventhandler.h>
40 #include <sys/sockio.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/mutex.h>
44 #include <sys/module.h>
45 #include <sys/kobj.h>
46 #include <sys/rman.h>
47 #include <sys/sbuf.h>
48 #include <sys/smp.h>
49 #include <sys/socket.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/taskqueue.h>
53 #include <sys/limits.h>
54 
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_types.h>
59 #include <net/if_media.h>
60 #include <net/bpf.h>
61 #include <net/ethernet.h>
62 #include <net/mp_ring.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/tcp_lro.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/if_ether.h>
70 #include <netinet/ip.h>
71 #include <netinet/ip6.h>
72 #include <netinet/tcp.h>
73 #include <netinet/ip_var.h>
74 #include <netinet6/ip6_var.h>
75 
76 #include <machine/bus.h>
77 #include <machine/in_cksum.h>
78 
79 #include <vm/vm.h>
80 #include <vm/pmap.h>
81 
82 #include <dev/led/led.h>
83 #include <dev/pci/pcireg.h>
84 #include <dev/pci/pcivar.h>
85 #include <dev/pci/pci_private.h>
86 
87 #include <net/iflib.h>
88 
89 #include "ifdi_if.h"
90 
91 #if defined(__i386__) || defined(__amd64__)
92 #include <sys/memdesc.h>
93 #include <machine/bus.h>
94 #include <machine/md_var.h>
95 #include <machine/specialreg.h>
96 #include <x86/include/busdma_impl.h>
97 #include <x86/iommu/busdma_dmar.h>
98 #endif
99 
100 #include <sys/bitstring.h>
101 /*
102  * enable accounting of every mbuf as it comes in to and goes out of
103  * iflib's software descriptor references
104  */
105 #define MEMORY_LOGGING 0
106 /*
107  * Enable mbuf vectors for compressing long mbuf chains
108  */
109 
110 /*
111  * NB:
112  * - Prefetching in tx cleaning should perhaps be a tunable. The distance ahead
113  *   we prefetch needs to be determined by the time spent in m_free vis a vis
114  *   the cost of a prefetch. This will of course vary based on the workload:
115  *      - NFLX's m_free path is dominated by vm-based M_EXT manipulation which
116  *        is quite expensive, thus suggesting very little prefetch.
117  *      - small packet forwarding which is just returning a single mbuf to
118  *        UMA will typically be very fast vis a vis the cost of a memory
119  *        access.
120  */
121 
122 
123 /*
124  * File organization:
125  *  - private structures
126  *  - iflib private utility functions
127  *  - ifnet functions
128  *  - vlan registry and other exported functions
129  *  - iflib public core functions
130  *
131  *
132  */
133 static MALLOC_DEFINE(M_IFLIB, "iflib", "ifnet library");
134 
135 struct iflib_txq;
136 typedef struct iflib_txq *iflib_txq_t;
137 struct iflib_rxq;
138 typedef struct iflib_rxq *iflib_rxq_t;
139 struct iflib_fl;
140 typedef struct iflib_fl *iflib_fl_t;
141 
142 struct iflib_ctx;
143 
144 static void iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid);
145 
146 typedef struct iflib_filter_info {
147 	driver_filter_t *ifi_filter;
148 	void *ifi_filter_arg;
149 	struct grouptask *ifi_task;
150 	void *ifi_ctx;
151 } *iflib_filter_info_t;
152 
153 struct iflib_ctx {
154 	KOBJ_FIELDS;
155    /*
156    * Pointer to hardware driver's softc
157    */
158 	void *ifc_softc;
159 	device_t ifc_dev;
160 	if_t ifc_ifp;
161 
162 	cpuset_t ifc_cpus;
163 	if_shared_ctx_t ifc_sctx;
164 	struct if_softc_ctx ifc_softc_ctx;
165 
166 	struct mtx ifc_ctx_mtx;
167 	struct mtx ifc_state_mtx;
168 
169 	uint16_t ifc_nhwtxqs;
170 
171 	iflib_txq_t ifc_txqs;
172 	iflib_rxq_t ifc_rxqs;
173 	uint32_t ifc_if_flags;
174 	uint32_t ifc_flags;
175 	uint32_t ifc_max_fl_buf_size;
176 	int ifc_in_detach;
177 
178 	int ifc_link_state;
179 	int ifc_link_irq;
180 	int ifc_watchdog_events;
181 	struct cdev *ifc_led_dev;
182 	struct resource *ifc_msix_mem;
183 
184 	struct if_irq ifc_legacy_irq;
185 	struct grouptask ifc_admin_task;
186 	struct grouptask ifc_vflr_task;
187 	struct iflib_filter_info ifc_filter_info;
188 	struct ifmedia	ifc_media;
189 
190 	struct sysctl_oid *ifc_sysctl_node;
191 	uint16_t ifc_sysctl_ntxqs;
192 	uint16_t ifc_sysctl_nrxqs;
193 	uint16_t ifc_sysctl_qs_eq_override;
194 	uint16_t ifc_sysctl_rx_budget;
195 
196 	qidx_t ifc_sysctl_ntxds[8];
197 	qidx_t ifc_sysctl_nrxds[8];
198 	struct if_txrx ifc_txrx;
199 #define isc_txd_encap  ifc_txrx.ift_txd_encap
200 #define isc_txd_flush  ifc_txrx.ift_txd_flush
201 #define isc_txd_credits_update  ifc_txrx.ift_txd_credits_update
202 #define isc_rxd_available ifc_txrx.ift_rxd_available
203 #define isc_rxd_pkt_get ifc_txrx.ift_rxd_pkt_get
204 #define isc_rxd_refill ifc_txrx.ift_rxd_refill
205 #define isc_rxd_flush ifc_txrx.ift_rxd_flush
206 #define isc_rxd_refill ifc_txrx.ift_rxd_refill
207 #define isc_rxd_refill ifc_txrx.ift_rxd_refill
208 #define isc_legacy_intr ifc_txrx.ift_legacy_intr
209 	eventhandler_tag ifc_vlan_attach_event;
210 	eventhandler_tag ifc_vlan_detach_event;
211 	uint8_t ifc_mac[ETHER_ADDR_LEN];
212 	char ifc_mtx_name[16];
213 };
214 
215 
216 void *
217 iflib_get_softc(if_ctx_t ctx)
218 {
219 
220 	return (ctx->ifc_softc);
221 }
222 
223 device_t
224 iflib_get_dev(if_ctx_t ctx)
225 {
226 
227 	return (ctx->ifc_dev);
228 }
229 
230 if_t
231 iflib_get_ifp(if_ctx_t ctx)
232 {
233 
234 	return (ctx->ifc_ifp);
235 }
236 
237 struct ifmedia *
238 iflib_get_media(if_ctx_t ctx)
239 {
240 
241 	return (&ctx->ifc_media);
242 }
243 
244 void
245 iflib_set_mac(if_ctx_t ctx, uint8_t mac[ETHER_ADDR_LEN])
246 {
247 
248 	bcopy(mac, ctx->ifc_mac, ETHER_ADDR_LEN);
249 }
250 
251 if_softc_ctx_t
252 iflib_get_softc_ctx(if_ctx_t ctx)
253 {
254 
255 	return (&ctx->ifc_softc_ctx);
256 }
257 
258 if_shared_ctx_t
259 iflib_get_sctx(if_ctx_t ctx)
260 {
261 
262 	return (ctx->ifc_sctx);
263 }
264 
265 #define IP_ALIGNED(m) ((((uintptr_t)(m)->m_data) & 0x3) == 0x2)
266 #define CACHE_PTR_INCREMENT (CACHE_LINE_SIZE/sizeof(void*))
267 #define CACHE_PTR_NEXT(ptr) ((void *)(((uintptr_t)(ptr)+CACHE_LINE_SIZE-1) & (CACHE_LINE_SIZE-1)))
268 
269 #define LINK_ACTIVE(ctx) ((ctx)->ifc_link_state == LINK_STATE_UP)
270 #define CTX_IS_VF(ctx) ((ctx)->ifc_sctx->isc_flags & IFLIB_IS_VF)
271 
272 #define RX_SW_DESC_MAP_CREATED	(1 << 0)
273 #define TX_SW_DESC_MAP_CREATED	(1 << 1)
274 #define RX_SW_DESC_INUSE        (1 << 3)
275 #define TX_SW_DESC_MAPPED       (1 << 4)
276 
277 #define	M_TOOBIG		M_PROTO1
278 
279 typedef struct iflib_sw_rx_desc_array {
280 	bus_dmamap_t	*ifsd_map;         /* bus_dma maps for packet */
281 	struct mbuf	**ifsd_m;           /* pkthdr mbufs */
282 	caddr_t		*ifsd_cl;          /* direct cluster pointer for rx */
283 	uint8_t		*ifsd_flags;
284 } iflib_rxsd_array_t;
285 
286 typedef struct iflib_sw_tx_desc_array {
287 	bus_dmamap_t    *ifsd_map;         /* bus_dma maps for packet */
288 	struct mbuf    **ifsd_m;           /* pkthdr mbufs */
289 	uint8_t		*ifsd_flags;
290 } if_txsd_vec_t;
291 
292 
293 /* magic number that should be high enough for any hardware */
294 #define IFLIB_MAX_TX_SEGS		128
295 /* bnxt supports 64 with hardware LRO enabled */
296 #define IFLIB_MAX_RX_SEGS		64
297 #define IFLIB_RX_COPY_THRESH		128
298 #define IFLIB_MAX_RX_REFRESH		32
299 /* The minimum descriptors per second before we start coalescing */
300 #define IFLIB_MIN_DESC_SEC		16384
301 #define IFLIB_DEFAULT_TX_UPDATE_FREQ	16
302 #define IFLIB_QUEUE_IDLE		0
303 #define IFLIB_QUEUE_HUNG		1
304 #define IFLIB_QUEUE_WORKING		2
305 /* maximum number of txqs that can share an rx interrupt */
306 #define IFLIB_MAX_TX_SHARED_INTR	4
307 
308 /* this should really scale with ring size - this is a fairly arbitrary value */
309 #define TX_BATCH_SIZE			32
310 
311 #define IFLIB_RESTART_BUDGET		8
312 
313 #define	IFC_LEGACY		0x001
314 #define	IFC_QFLUSH		0x002
315 #define	IFC_MULTISEG		0x004
316 #define	IFC_DMAR		0x008
317 #define	IFC_SC_ALLOCATED	0x010
318 #define	IFC_INIT_DONE		0x020
319 #define	IFC_PREFETCH		0x040
320 #define	IFC_DO_RESET		0x080
321 #define	IFC_DO_WATCHDOG		0x100
322 #define	IFC_CHECK_HUNG		0x200
323 
324 
325 #define CSUM_OFFLOAD		(CSUM_IP_TSO|CSUM_IP6_TSO|CSUM_IP| \
326 				 CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP| \
327 				 CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP)
328 struct iflib_txq {
329 	qidx_t		ift_in_use;
330 	qidx_t		ift_cidx;
331 	qidx_t		ift_cidx_processed;
332 	qidx_t		ift_pidx;
333 	uint8_t		ift_gen;
334 	uint8_t		ift_br_offset;
335 	uint16_t	ift_npending;
336 	uint16_t	ift_db_pending;
337 	uint16_t	ift_rs_pending;
338 	/* implicit pad */
339 	uint8_t		ift_txd_size[8];
340 	uint64_t	ift_processed;
341 	uint64_t	ift_cleaned;
342 	uint64_t	ift_cleaned_prev;
343 #if MEMORY_LOGGING
344 	uint64_t	ift_enqueued;
345 	uint64_t	ift_dequeued;
346 #endif
347 	uint64_t	ift_no_tx_dma_setup;
348 	uint64_t	ift_no_desc_avail;
349 	uint64_t	ift_mbuf_defrag_failed;
350 	uint64_t	ift_mbuf_defrag;
351 	uint64_t	ift_map_failed;
352 	uint64_t	ift_txd_encap_efbig;
353 	uint64_t	ift_pullups;
354 
355 	struct mtx	ift_mtx;
356 	struct mtx	ift_db_mtx;
357 
358 	/* constant values */
359 	if_ctx_t	ift_ctx;
360 	struct ifmp_ring        *ift_br;
361 	struct grouptask	ift_task;
362 	qidx_t		ift_size;
363 	uint16_t	ift_id;
364 	struct callout	ift_timer;
365 
366 	if_txsd_vec_t	ift_sds;
367 	uint8_t		ift_qstatus;
368 	uint8_t		ift_closed;
369 	uint8_t		ift_update_freq;
370 	struct iflib_filter_info ift_filter_info;
371 	bus_dma_tag_t		ift_desc_tag;
372 	bus_dma_tag_t		ift_tso_desc_tag;
373 	iflib_dma_info_t	ift_ifdi;
374 #define MTX_NAME_LEN 16
375 	char                    ift_mtx_name[MTX_NAME_LEN];
376 	char                    ift_db_mtx_name[MTX_NAME_LEN];
377 	bus_dma_segment_t	ift_segs[IFLIB_MAX_TX_SEGS]  __aligned(CACHE_LINE_SIZE);
378 #ifdef IFLIB_DIAGNOSTICS
379 	uint64_t ift_cpu_exec_count[256];
380 #endif
381 } __aligned(CACHE_LINE_SIZE);
382 
383 struct iflib_fl {
384 	qidx_t		ifl_cidx;
385 	qidx_t		ifl_pidx;
386 	qidx_t		ifl_credits;
387 	uint8_t		ifl_gen;
388 	uint8_t		ifl_rxd_size;
389 #if MEMORY_LOGGING
390 	uint64_t	ifl_m_enqueued;
391 	uint64_t	ifl_m_dequeued;
392 	uint64_t	ifl_cl_enqueued;
393 	uint64_t	ifl_cl_dequeued;
394 #endif
395 	/* implicit pad */
396 
397 	bitstr_t 	*ifl_rx_bitmap;
398 	qidx_t		ifl_fragidx;
399 	/* constant */
400 	qidx_t		ifl_size;
401 	uint16_t	ifl_buf_size;
402 	uint16_t	ifl_cltype;
403 	uma_zone_t	ifl_zone;
404 	iflib_rxsd_array_t	ifl_sds;
405 	iflib_rxq_t	ifl_rxq;
406 	uint8_t		ifl_id;
407 	bus_dma_tag_t           ifl_desc_tag;
408 	iflib_dma_info_t	ifl_ifdi;
409 	uint64_t	ifl_bus_addrs[IFLIB_MAX_RX_REFRESH] __aligned(CACHE_LINE_SIZE);
410 	caddr_t		ifl_vm_addrs[IFLIB_MAX_RX_REFRESH];
411 	qidx_t	ifl_rxd_idxs[IFLIB_MAX_RX_REFRESH];
412 }  __aligned(CACHE_LINE_SIZE);
413 
414 static inline qidx_t
415 get_inuse(int size, qidx_t cidx, qidx_t pidx, uint8_t gen)
416 {
417 	qidx_t used;
418 
419 	if (pidx > cidx)
420 		used = pidx - cidx;
421 	else if (pidx < cidx)
422 		used = size - cidx + pidx;
423 	else if (gen == 0 && pidx == cidx)
424 		used = 0;
425 	else if (gen == 1 && pidx == cidx)
426 		used = size;
427 	else
428 		panic("bad state");
429 
430 	return (used);
431 }
432 
433 #define TXQ_AVAIL(txq) (txq->ift_size - get_inuse(txq->ift_size, txq->ift_cidx, txq->ift_pidx, txq->ift_gen))
434 
435 #define IDXDIFF(head, tail, wrap) \
436 	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
437 
438 struct iflib_rxq {
439 	/* If there is a separate completion queue -
440 	 * these are the cq cidx and pidx. Otherwise
441 	 * these are unused.
442 	 */
443 	qidx_t		ifr_size;
444 	qidx_t		ifr_cq_cidx;
445 	qidx_t		ifr_cq_pidx;
446 	uint8_t		ifr_cq_gen;
447 	uint8_t		ifr_fl_offset;
448 
449 	if_ctx_t	ifr_ctx;
450 	iflib_fl_t	ifr_fl;
451 	uint64_t	ifr_rx_irq;
452 	uint16_t	ifr_id;
453 	uint8_t		ifr_lro_enabled;
454 	uint8_t		ifr_nfl;
455 	uint8_t		ifr_ntxqirq;
456 	uint8_t		ifr_txqid[IFLIB_MAX_TX_SHARED_INTR];
457 	struct lro_ctrl			ifr_lc;
458 	struct grouptask        ifr_task;
459 	struct iflib_filter_info ifr_filter_info;
460 	iflib_dma_info_t		ifr_ifdi;
461 
462 	/* dynamically allocate if any drivers need a value substantially larger than this */
463 	struct if_rxd_frag	ifr_frags[IFLIB_MAX_RX_SEGS] __aligned(CACHE_LINE_SIZE);
464 #ifdef IFLIB_DIAGNOSTICS
465 	uint64_t ifr_cpu_exec_count[256];
466 #endif
467 }  __aligned(CACHE_LINE_SIZE);
468 
469 typedef struct if_rxsd {
470 	caddr_t *ifsd_cl;
471 	struct mbuf **ifsd_m;
472 	iflib_fl_t ifsd_fl;
473 	qidx_t ifsd_cidx;
474 } *if_rxsd_t;
475 
476 /* multiple of word size */
477 #ifdef __LP64__
478 #define PKT_INFO_SIZE	6
479 #define RXD_INFO_SIZE	5
480 #define PKT_TYPE uint64_t
481 #else
482 #define PKT_INFO_SIZE	11
483 #define RXD_INFO_SIZE	8
484 #define PKT_TYPE uint32_t
485 #endif
486 #define PKT_LOOP_BOUND  ((PKT_INFO_SIZE/3)*3)
487 #define RXD_LOOP_BOUND  ((RXD_INFO_SIZE/4)*4)
488 
489 typedef struct if_pkt_info_pad {
490 	PKT_TYPE pkt_val[PKT_INFO_SIZE];
491 } *if_pkt_info_pad_t;
492 typedef struct if_rxd_info_pad {
493 	PKT_TYPE rxd_val[RXD_INFO_SIZE];
494 } *if_rxd_info_pad_t;
495 
496 CTASSERT(sizeof(struct if_pkt_info_pad) == sizeof(struct if_pkt_info));
497 CTASSERT(sizeof(struct if_rxd_info_pad) == sizeof(struct if_rxd_info));
498 
499 
500 static inline void
501 pkt_info_zero(if_pkt_info_t pi)
502 {
503 	if_pkt_info_pad_t pi_pad;
504 
505 	pi_pad = (if_pkt_info_pad_t)pi;
506 	pi_pad->pkt_val[0] = 0; pi_pad->pkt_val[1] = 0; pi_pad->pkt_val[2] = 0;
507 	pi_pad->pkt_val[3] = 0; pi_pad->pkt_val[4] = 0; pi_pad->pkt_val[5] = 0;
508 #ifndef __LP64__
509 	pi_pad->pkt_val[6] = 0; pi_pad->pkt_val[7] = 0; pi_pad->pkt_val[8] = 0;
510 	pi_pad->pkt_val[9] = 0; pi_pad->pkt_val[10] = 0;
511 #endif
512 }
513 
514 static inline void
515 rxd_info_zero(if_rxd_info_t ri)
516 {
517 	if_rxd_info_pad_t ri_pad;
518 	int i;
519 
520 	ri_pad = (if_rxd_info_pad_t)ri;
521 	for (i = 0; i < RXD_LOOP_BOUND; i += 4) {
522 		ri_pad->rxd_val[i] = 0;
523 		ri_pad->rxd_val[i+1] = 0;
524 		ri_pad->rxd_val[i+2] = 0;
525 		ri_pad->rxd_val[i+3] = 0;
526 	}
527 #ifdef __LP64__
528 	ri_pad->rxd_val[RXD_INFO_SIZE-1] = 0;
529 #endif
530 }
531 
532 /*
533  * Only allow a single packet to take up most 1/nth of the tx ring
534  */
535 #define MAX_SINGLE_PACKET_FRACTION 12
536 #define IF_BAD_DMA (bus_addr_t)-1
537 
538 #define CTX_ACTIVE(ctx) ((if_getdrvflags((ctx)->ifc_ifp) & IFF_DRV_RUNNING))
539 
540 #define CTX_LOCK_INIT(_sc, _name)  mtx_init(&(_sc)->ifc_ctx_mtx, _name, "iflib ctx lock", MTX_DEF)
541 #define CTX_LOCK(ctx) mtx_lock(&(ctx)->ifc_ctx_mtx)
542 #define CTX_UNLOCK(ctx) mtx_unlock(&(ctx)->ifc_ctx_mtx)
543 #define CTX_LOCK_DESTROY(ctx) mtx_destroy(&(ctx)->ifc_ctx_mtx)
544 
545 
546 #define STATE_LOCK_INIT(_sc, _name)  mtx_init(&(_sc)->ifc_state_mtx, _name, "iflib state lock", MTX_DEF)
547 #define STATE_LOCK(ctx) mtx_lock(&(ctx)->ifc_state_mtx)
548 #define STATE_UNLOCK(ctx) mtx_unlock(&(ctx)->ifc_state_mtx)
549 #define STATE_LOCK_DESTROY(ctx) mtx_destroy(&(ctx)->ifc_state_mtx)
550 
551 
552 
553 #define CALLOUT_LOCK(txq)	mtx_lock(&txq->ift_mtx)
554 #define CALLOUT_UNLOCK(txq) 	mtx_unlock(&txq->ift_mtx)
555 
556 
557 /* Our boot-time initialization hook */
558 static int	iflib_module_event_handler(module_t, int, void *);
559 
560 static moduledata_t iflib_moduledata = {
561 	"iflib",
562 	iflib_module_event_handler,
563 	NULL
564 };
565 
566 DECLARE_MODULE(iflib, iflib_moduledata, SI_SUB_INIT_IF, SI_ORDER_ANY);
567 MODULE_VERSION(iflib, 1);
568 
569 MODULE_DEPEND(iflib, pci, 1, 1, 1);
570 MODULE_DEPEND(iflib, ether, 1, 1, 1);
571 
572 TASKQGROUP_DEFINE(if_io_tqg, mp_ncpus, 1);
573 TASKQGROUP_DEFINE(if_config_tqg, 1, 1);
574 
575 #ifndef IFLIB_DEBUG_COUNTERS
576 #ifdef INVARIANTS
577 #define IFLIB_DEBUG_COUNTERS 1
578 #else
579 #define IFLIB_DEBUG_COUNTERS 0
580 #endif /* !INVARIANTS */
581 #endif
582 
583 static SYSCTL_NODE(_net, OID_AUTO, iflib, CTLFLAG_RD, 0,
584                    "iflib driver parameters");
585 
586 /*
587  * XXX need to ensure that this can't accidentally cause the head to be moved backwards
588  */
589 static int iflib_min_tx_latency = 0;
590 SYSCTL_INT(_net_iflib, OID_AUTO, min_tx_latency, CTLFLAG_RW,
591 		   &iflib_min_tx_latency, 0, "minimize transmit latency at the possible expense of throughput");
592 static int iflib_no_tx_batch = 0;
593 SYSCTL_INT(_net_iflib, OID_AUTO, no_tx_batch, CTLFLAG_RW,
594 		   &iflib_no_tx_batch, 0, "minimize transmit latency at the possible expense of throughput");
595 
596 
597 #if IFLIB_DEBUG_COUNTERS
598 
599 static int iflib_tx_seen;
600 static int iflib_tx_sent;
601 static int iflib_tx_encap;
602 static int iflib_rx_allocs;
603 static int iflib_fl_refills;
604 static int iflib_fl_refills_large;
605 static int iflib_tx_frees;
606 
607 SYSCTL_INT(_net_iflib, OID_AUTO, tx_seen, CTLFLAG_RD,
608 		   &iflib_tx_seen, 0, "# tx mbufs seen");
609 SYSCTL_INT(_net_iflib, OID_AUTO, tx_sent, CTLFLAG_RD,
610 		   &iflib_tx_sent, 0, "# tx mbufs sent");
611 SYSCTL_INT(_net_iflib, OID_AUTO, tx_encap, CTLFLAG_RD,
612 		   &iflib_tx_encap, 0, "# tx mbufs encapped");
613 SYSCTL_INT(_net_iflib, OID_AUTO, tx_frees, CTLFLAG_RD,
614 		   &iflib_tx_frees, 0, "# tx frees");
615 SYSCTL_INT(_net_iflib, OID_AUTO, rx_allocs, CTLFLAG_RD,
616 		   &iflib_rx_allocs, 0, "# rx allocations");
617 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills, CTLFLAG_RD,
618 		   &iflib_fl_refills, 0, "# refills");
619 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills_large, CTLFLAG_RD,
620 		   &iflib_fl_refills_large, 0, "# large refills");
621 
622 
623 static int iflib_txq_drain_flushing;
624 static int iflib_txq_drain_oactive;
625 static int iflib_txq_drain_notready;
626 static int iflib_txq_drain_encapfail;
627 
628 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_flushing, CTLFLAG_RD,
629 		   &iflib_txq_drain_flushing, 0, "# drain flushes");
630 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_oactive, CTLFLAG_RD,
631 		   &iflib_txq_drain_oactive, 0, "# drain oactives");
632 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_notready, CTLFLAG_RD,
633 		   &iflib_txq_drain_notready, 0, "# drain notready");
634 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_encapfail, CTLFLAG_RD,
635 		   &iflib_txq_drain_encapfail, 0, "# drain encap fails");
636 
637 
638 static int iflib_encap_load_mbuf_fail;
639 static int iflib_encap_pad_mbuf_fail;
640 static int iflib_encap_txq_avail_fail;
641 static int iflib_encap_txd_encap_fail;
642 
643 SYSCTL_INT(_net_iflib, OID_AUTO, encap_load_mbuf_fail, CTLFLAG_RD,
644 		   &iflib_encap_load_mbuf_fail, 0, "# busdma load failures");
645 SYSCTL_INT(_net_iflib, OID_AUTO, encap_pad_mbuf_fail, CTLFLAG_RD,
646 		   &iflib_encap_pad_mbuf_fail, 0, "# runt frame pad failures");
647 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txq_avail_fail, CTLFLAG_RD,
648 		   &iflib_encap_txq_avail_fail, 0, "# txq avail failures");
649 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txd_encap_fail, CTLFLAG_RD,
650 		   &iflib_encap_txd_encap_fail, 0, "# driver encap failures");
651 
652 static int iflib_task_fn_rxs;
653 static int iflib_rx_intr_enables;
654 static int iflib_fast_intrs;
655 static int iflib_intr_link;
656 static int iflib_intr_msix;
657 static int iflib_rx_unavail;
658 static int iflib_rx_ctx_inactive;
659 static int iflib_rx_zero_len;
660 static int iflib_rx_if_input;
661 static int iflib_rx_mbuf_null;
662 static int iflib_rxd_flush;
663 
664 static int iflib_verbose_debug;
665 
666 SYSCTL_INT(_net_iflib, OID_AUTO, intr_link, CTLFLAG_RD,
667 		   &iflib_intr_link, 0, "# intr link calls");
668 SYSCTL_INT(_net_iflib, OID_AUTO, intr_msix, CTLFLAG_RD,
669 		   &iflib_intr_msix, 0, "# intr msix calls");
670 SYSCTL_INT(_net_iflib, OID_AUTO, task_fn_rx, CTLFLAG_RD,
671 		   &iflib_task_fn_rxs, 0, "# task_fn_rx calls");
672 SYSCTL_INT(_net_iflib, OID_AUTO, rx_intr_enables, CTLFLAG_RD,
673 		   &iflib_rx_intr_enables, 0, "# rx intr enables");
674 SYSCTL_INT(_net_iflib, OID_AUTO, fast_intrs, CTLFLAG_RD,
675 		   &iflib_fast_intrs, 0, "# fast_intr calls");
676 SYSCTL_INT(_net_iflib, OID_AUTO, rx_unavail, CTLFLAG_RD,
677 		   &iflib_rx_unavail, 0, "# times rxeof called with no available data");
678 SYSCTL_INT(_net_iflib, OID_AUTO, rx_ctx_inactive, CTLFLAG_RD,
679 		   &iflib_rx_ctx_inactive, 0, "# times rxeof called with inactive context");
680 SYSCTL_INT(_net_iflib, OID_AUTO, rx_zero_len, CTLFLAG_RD,
681 		   &iflib_rx_zero_len, 0, "# times rxeof saw zero len mbuf");
682 SYSCTL_INT(_net_iflib, OID_AUTO, rx_if_input, CTLFLAG_RD,
683 		   &iflib_rx_if_input, 0, "# times rxeof called if_input");
684 SYSCTL_INT(_net_iflib, OID_AUTO, rx_mbuf_null, CTLFLAG_RD,
685 		   &iflib_rx_mbuf_null, 0, "# times rxeof got null mbuf");
686 SYSCTL_INT(_net_iflib, OID_AUTO, rxd_flush, CTLFLAG_RD,
687 	         &iflib_rxd_flush, 0, "# times rxd_flush called");
688 SYSCTL_INT(_net_iflib, OID_AUTO, verbose_debug, CTLFLAG_RW,
689 		   &iflib_verbose_debug, 0, "enable verbose debugging");
690 
691 #define DBG_COUNTER_INC(name) atomic_add_int(&(iflib_ ## name), 1)
692 static void
693 iflib_debug_reset(void)
694 {
695 	iflib_tx_seen = iflib_tx_sent = iflib_tx_encap = iflib_rx_allocs =
696 		iflib_fl_refills = iflib_fl_refills_large = iflib_tx_frees =
697 		iflib_txq_drain_flushing = iflib_txq_drain_oactive =
698 		iflib_txq_drain_notready = iflib_txq_drain_encapfail =
699 		iflib_encap_load_mbuf_fail = iflib_encap_pad_mbuf_fail =
700 		iflib_encap_txq_avail_fail = iflib_encap_txd_encap_fail =
701 		iflib_task_fn_rxs = iflib_rx_intr_enables = iflib_fast_intrs =
702 		iflib_intr_link = iflib_intr_msix = iflib_rx_unavail =
703 		iflib_rx_ctx_inactive = iflib_rx_zero_len = iflib_rx_if_input =
704 		iflib_rx_mbuf_null = iflib_rxd_flush = 0;
705 }
706 
707 #else
708 #define DBG_COUNTER_INC(name)
709 static void iflib_debug_reset(void) {}
710 #endif
711 
712 
713 
714 #define IFLIB_DEBUG 0
715 
716 static void iflib_tx_structures_free(if_ctx_t ctx);
717 static void iflib_rx_structures_free(if_ctx_t ctx);
718 static int iflib_queues_alloc(if_ctx_t ctx);
719 static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq);
720 static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget);
721 static int iflib_qset_structures_setup(if_ctx_t ctx);
722 static int iflib_msix_init(if_ctx_t ctx);
723 static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filterarg, int *rid, char *str);
724 static void iflib_txq_check_drain(iflib_txq_t txq, int budget);
725 static uint32_t iflib_txq_can_drain(struct ifmp_ring *);
726 static int iflib_register(if_ctx_t);
727 static void iflib_init_locked(if_ctx_t ctx);
728 static void iflib_add_device_sysctl_pre(if_ctx_t ctx);
729 static void iflib_add_device_sysctl_post(if_ctx_t ctx);
730 static void iflib_ifmp_purge(iflib_txq_t txq);
731 static void _iflib_pre_assert(if_softc_ctx_t scctx);
732 static void iflib_stop(if_ctx_t ctx);
733 static void iflib_if_init_locked(if_ctx_t ctx);
734 #ifndef __NO_STRICT_ALIGNMENT
735 static struct mbuf * iflib_fixup_rx(struct mbuf *m);
736 #endif
737 
738 #ifdef DEV_NETMAP
739 #include <sys/selinfo.h>
740 #include <net/netmap.h>
741 #include <dev/netmap/netmap_kern.h>
742 
743 MODULE_DEPEND(iflib, netmap, 1, 1, 1);
744 
745 static int netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, uint32_t nm_i, bool init);
746 
747 /*
748  * device-specific sysctl variables:
749  *
750  * iflib_crcstrip: 0: keep CRC in rx frames (default), 1: strip it.
751  *	During regular operations the CRC is stripped, but on some
752  *	hardware reception of frames not multiple of 64 is slower,
753  *	so using crcstrip=0 helps in benchmarks.
754  *
755  * iflib_rx_miss, iflib_rx_miss_bufs:
756  *	count packets that might be missed due to lost interrupts.
757  */
758 SYSCTL_DECL(_dev_netmap);
759 /*
760  * The xl driver by default strips CRCs and we do not override it.
761  */
762 
763 int iflib_crcstrip = 1;
764 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_crcstrip,
765     CTLFLAG_RW, &iflib_crcstrip, 1, "strip CRC on rx frames");
766 
767 int iflib_rx_miss, iflib_rx_miss_bufs;
768 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss,
769     CTLFLAG_RW, &iflib_rx_miss, 0, "potentially missed rx intr");
770 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss_bufs,
771     CTLFLAG_RW, &iflib_rx_miss_bufs, 0, "potentially missed rx intr bufs");
772 
773 /*
774  * Register/unregister. We are already under netmap lock.
775  * Only called on the first register or the last unregister.
776  */
777 static int
778 iflib_netmap_register(struct netmap_adapter *na, int onoff)
779 {
780 	struct ifnet *ifp = na->ifp;
781 	if_ctx_t ctx = ifp->if_softc;
782 	int status;
783 
784 	CTX_LOCK(ctx);
785 	IFDI_INTR_DISABLE(ctx);
786 
787 	/* Tell the stack that the interface is no longer active */
788 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
789 
790 	if (!CTX_IS_VF(ctx))
791 		IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip);
792 
793 	/* enable or disable flags and callbacks in na and ifp */
794 	if (onoff) {
795 		nm_set_native_flags(na);
796 	} else {
797 		nm_clear_native_flags(na);
798 	}
799 	iflib_stop(ctx);
800 	iflib_init_locked(ctx);
801 	IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); // XXX why twice ?
802 	status = ifp->if_drv_flags & IFF_DRV_RUNNING ? 0 : 1;
803 	if (status)
804 		nm_clear_native_flags(na);
805 	CTX_UNLOCK(ctx);
806 	return (status);
807 }
808 
809 static int
810 netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, uint32_t nm_i, bool init)
811 {
812 	struct netmap_adapter *na = kring->na;
813 	u_int const lim = kring->nkr_num_slots - 1;
814 	u_int head = kring->rhead;
815 	struct netmap_ring *ring = kring->ring;
816 	bus_dmamap_t *map;
817 	struct if_rxd_update iru;
818 	if_ctx_t ctx = rxq->ifr_ctx;
819 	iflib_fl_t fl = &rxq->ifr_fl[0];
820 	uint32_t refill_pidx, nic_i;
821 
822 	if (nm_i == head && __predict_true(!init))
823 		return 0;
824 	iru_init(&iru, rxq, 0 /* flid */);
825 	map = fl->ifl_sds.ifsd_map;
826 	refill_pidx = netmap_idx_k2n(kring, nm_i);
827 	/*
828 	 * IMPORTANT: we must leave one free slot in the ring,
829 	 * so move head back by one unit
830 	 */
831 	head = nm_prev(head, lim);
832 	while (nm_i != head) {
833 		for (int tmp_pidx = 0; tmp_pidx < IFLIB_MAX_RX_REFRESH && nm_i != head; tmp_pidx++) {
834 			struct netmap_slot *slot = &ring->slot[nm_i];
835 			void *addr = PNMB(na, slot, &fl->ifl_bus_addrs[tmp_pidx]);
836 			uint32_t nic_i_dma = refill_pidx;
837 			nic_i = netmap_idx_k2n(kring, nm_i);
838 
839 			MPASS(tmp_pidx < IFLIB_MAX_RX_REFRESH);
840 
841 			if (addr == NETMAP_BUF_BASE(na)) /* bad buf */
842 			        return netmap_ring_reinit(kring);
843 
844 			fl->ifl_vm_addrs[tmp_pidx] = addr;
845 			if (__predict_false(init) && map) {
846 				netmap_load_map(na, fl->ifl_ifdi->idi_tag, map[nic_i], addr);
847 			} else if (map && (slot->flags & NS_BUF_CHANGED)) {
848 				/* buffer has changed, reload map */
849 				netmap_reload_map(na, fl->ifl_ifdi->idi_tag, map[nic_i], addr);
850 			}
851 			slot->flags &= ~NS_BUF_CHANGED;
852 
853 			nm_i = nm_next(nm_i, lim);
854 			fl->ifl_rxd_idxs[tmp_pidx] = nic_i = nm_next(nic_i, lim);
855 			if (nm_i != head && tmp_pidx < IFLIB_MAX_RX_REFRESH-1)
856 				continue;
857 
858 			iru.iru_pidx = refill_pidx;
859 			iru.iru_count = tmp_pidx+1;
860 			ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
861 
862 			refill_pidx = nic_i;
863 			if (map == NULL)
864 				continue;
865 
866 			for (int n = 0; n < iru.iru_count; n++) {
867 				bus_dmamap_sync(fl->ifl_ifdi->idi_tag, map[nic_i_dma],
868 						BUS_DMASYNC_PREREAD);
869 				/* XXX - change this to not use the netmap func*/
870 				nic_i_dma = nm_next(nic_i_dma, lim);
871 			}
872 		}
873 	}
874 	kring->nr_hwcur = head;
875 
876 	if (map)
877 		bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
878 				BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
879 	ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, fl->ifl_id, nic_i);
880 	return (0);
881 }
882 
883 /*
884  * Reconcile kernel and user view of the transmit ring.
885  *
886  * All information is in the kring.
887  * Userspace wants to send packets up to the one before kring->rhead,
888  * kernel knows kring->nr_hwcur is the first unsent packet.
889  *
890  * Here we push packets out (as many as possible), and possibly
891  * reclaim buffers from previously completed transmission.
892  *
893  * The caller (netmap) guarantees that there is only one instance
894  * running at any time. Any interference with other driver
895  * methods should be handled by the individual drivers.
896  */
897 static int
898 iflib_netmap_txsync(struct netmap_kring *kring, int flags)
899 {
900 	struct netmap_adapter *na = kring->na;
901 	struct ifnet *ifp = na->ifp;
902 	struct netmap_ring *ring = kring->ring;
903 	u_int nm_i;	/* index into the netmap ring */
904 	u_int nic_i;	/* index into the NIC ring */
905 	u_int n;
906 	u_int const lim = kring->nkr_num_slots - 1;
907 	u_int const head = kring->rhead;
908 	struct if_pkt_info pi;
909 
910 	/*
911 	 * interrupts on every tx packet are expensive so request
912 	 * them every half ring, or where NS_REPORT is set
913 	 */
914 	u_int report_frequency = kring->nkr_num_slots >> 1;
915 	/* device-specific */
916 	if_ctx_t ctx = ifp->if_softc;
917 	iflib_txq_t txq = &ctx->ifc_txqs[kring->ring_id];
918 
919 	if (txq->ift_sds.ifsd_map)
920 		bus_dmamap_sync(txq->ift_desc_tag, txq->ift_ifdi->idi_map,
921 				BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
922 
923 
924 	/*
925 	 * First part: process new packets to send.
926 	 * nm_i is the current index in the netmap ring,
927 	 * nic_i is the corresponding index in the NIC ring.
928 	 *
929 	 * If we have packets to send (nm_i != head)
930 	 * iterate over the netmap ring, fetch length and update
931 	 * the corresponding slot in the NIC ring. Some drivers also
932 	 * need to update the buffer's physical address in the NIC slot
933 	 * even NS_BUF_CHANGED is not set (PNMB computes the addresses).
934 	 *
935 	 * The netmap_reload_map() calls is especially expensive,
936 	 * even when (as in this case) the tag is 0, so do only
937 	 * when the buffer has actually changed.
938 	 *
939 	 * If possible do not set the report/intr bit on all slots,
940 	 * but only a few times per ring or when NS_REPORT is set.
941 	 *
942 	 * Finally, on 10G and faster drivers, it might be useful
943 	 * to prefetch the next slot and txr entry.
944 	 */
945 
946 	nm_i = netmap_idx_n2k(kring, kring->nr_hwcur);
947 	pkt_info_zero(&pi);
948 	pi.ipi_segs = txq->ift_segs;
949 	pi.ipi_qsidx = kring->ring_id;
950 	if (nm_i != head) {	/* we have new packets to send */
951 		nic_i = netmap_idx_k2n(kring, nm_i);
952 
953 		__builtin_prefetch(&ring->slot[nm_i]);
954 		__builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i]);
955 		if (txq->ift_sds.ifsd_map)
956 			__builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i]);
957 
958 		for (n = 0; nm_i != head; n++) {
959 			struct netmap_slot *slot = &ring->slot[nm_i];
960 			u_int len = slot->len;
961 			uint64_t paddr;
962 			void *addr = PNMB(na, slot, &paddr);
963 			int flags = (slot->flags & NS_REPORT ||
964 				nic_i == 0 || nic_i == report_frequency) ?
965 				IPI_TX_INTR : 0;
966 
967 			/* device-specific */
968 			pi.ipi_len = len;
969 			pi.ipi_segs[0].ds_addr = paddr;
970 			pi.ipi_segs[0].ds_len = len;
971 			pi.ipi_nsegs = 1;
972 			pi.ipi_ndescs = 0;
973 			pi.ipi_pidx = nic_i;
974 			pi.ipi_flags = flags;
975 
976 			/* Fill the slot in the NIC ring. */
977 			ctx->isc_txd_encap(ctx->ifc_softc, &pi);
978 
979 			/* prefetch for next round */
980 			__builtin_prefetch(&ring->slot[nm_i + 1]);
981 			__builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i + 1]);
982 			if (txq->ift_sds.ifsd_map) {
983 				__builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i + 1]);
984 
985 				NM_CHECK_ADDR_LEN(na, addr, len);
986 
987 				if (slot->flags & NS_BUF_CHANGED) {
988 					/* buffer has changed, reload map */
989 					netmap_reload_map(na, txq->ift_desc_tag, txq->ift_sds.ifsd_map[nic_i], addr);
990 				}
991 				/* make sure changes to the buffer are synced */
992 				bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_sds.ifsd_map[nic_i],
993 						BUS_DMASYNC_PREWRITE);
994 			}
995 			slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED);
996 			nm_i = nm_next(nm_i, lim);
997 			nic_i = nm_next(nic_i, lim);
998 		}
999 		kring->nr_hwcur = head;
1000 
1001 		/* synchronize the NIC ring */
1002 		if (txq->ift_sds.ifsd_map)
1003 			bus_dmamap_sync(txq->ift_desc_tag, txq->ift_ifdi->idi_map,
1004 						BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1005 
1006 		/* (re)start the tx unit up to slot nic_i (excluded) */
1007 		ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, nic_i);
1008 	}
1009 
1010 	/*
1011 	 * Second part: reclaim buffers for completed transmissions.
1012 	 */
1013 	if (iflib_tx_credits_update(ctx, txq)) {
1014 		/* some tx completed, increment avail */
1015 		nic_i = txq->ift_cidx_processed;
1016 		kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim);
1017 	}
1018 	return (0);
1019 }
1020 
1021 /*
1022  * Reconcile kernel and user view of the receive ring.
1023  * Same as for the txsync, this routine must be efficient.
1024  * The caller guarantees a single invocations, but races against
1025  * the rest of the driver should be handled here.
1026  *
1027  * On call, kring->rhead is the first packet that userspace wants
1028  * to keep, and kring->rcur is the wakeup point.
1029  * The kernel has previously reported packets up to kring->rtail.
1030  *
1031  * If (flags & NAF_FORCE_READ) also check for incoming packets irrespective
1032  * of whether or not we received an interrupt.
1033  */
1034 static int
1035 iflib_netmap_rxsync(struct netmap_kring *kring, int flags)
1036 {
1037 	struct netmap_adapter *na = kring->na;
1038 	struct netmap_ring *ring = kring->ring;
1039 	uint32_t nm_i;	/* index into the netmap ring */
1040 	uint32_t nic_i;	/* index into the NIC ring */
1041 	u_int i, n;
1042 	u_int const lim = kring->nkr_num_slots - 1;
1043 	u_int const head = netmap_idx_n2k(kring, kring->rhead);
1044 	int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR;
1045 	struct if_rxd_info ri;
1046 
1047 	struct ifnet *ifp = na->ifp;
1048 	if_ctx_t ctx = ifp->if_softc;
1049 	iflib_rxq_t rxq = &ctx->ifc_rxqs[kring->ring_id];
1050 	iflib_fl_t fl = rxq->ifr_fl;
1051 	if (head > lim)
1052 		return netmap_ring_reinit(kring);
1053 
1054 	/* XXX check sync modes */
1055 	for (i = 0, fl = rxq->ifr_fl; i < rxq->ifr_nfl; i++, fl++) {
1056 		if (fl->ifl_sds.ifsd_map == NULL)
1057 			continue;
1058 		bus_dmamap_sync(rxq->ifr_fl[i].ifl_desc_tag, fl->ifl_ifdi->idi_map,
1059 				BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1060 	}
1061 	/*
1062 	 * First part: import newly received packets.
1063 	 *
1064 	 * nm_i is the index of the next free slot in the netmap ring,
1065 	 * nic_i is the index of the next received packet in the NIC ring,
1066 	 * and they may differ in case if_init() has been called while
1067 	 * in netmap mode. For the receive ring we have
1068 	 *
1069 	 *	nic_i = rxr->next_check;
1070 	 *	nm_i = kring->nr_hwtail (previous)
1071 	 * and
1072 	 *	nm_i == (nic_i + kring->nkr_hwofs) % ring_size
1073 	 *
1074 	 * rxr->next_check is set to 0 on a ring reinit
1075 	 */
1076 	if (netmap_no_pendintr || force_update) {
1077 		int crclen = iflib_crcstrip ? 0 : 4;
1078 		int error, avail;
1079 
1080 		for (i = 0; i < rxq->ifr_nfl; i++) {
1081 			fl = &rxq->ifr_fl[i];
1082 			nic_i = fl->ifl_cidx;
1083 			nm_i = netmap_idx_n2k(kring, nic_i);
1084 			avail = iflib_rxd_avail(ctx, rxq, nic_i, USHRT_MAX);
1085 			for (n = 0; avail > 0; n++, avail--) {
1086 				rxd_info_zero(&ri);
1087 				ri.iri_frags = rxq->ifr_frags;
1088 				ri.iri_qsidx = kring->ring_id;
1089 				ri.iri_ifp = ctx->ifc_ifp;
1090 				ri.iri_cidx = nic_i;
1091 
1092 				error = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri);
1093 				ring->slot[nm_i].len = error ? 0 : ri.iri_len - crclen;
1094 				ring->slot[nm_i].flags = 0;
1095 				if (fl->ifl_sds.ifsd_map)
1096 					bus_dmamap_sync(fl->ifl_ifdi->idi_tag,
1097 							fl->ifl_sds.ifsd_map[nic_i], BUS_DMASYNC_POSTREAD);
1098 				nm_i = nm_next(nm_i, lim);
1099 				nic_i = nm_next(nic_i, lim);
1100 			}
1101 			if (n) { /* update the state variables */
1102 				if (netmap_no_pendintr && !force_update) {
1103 					/* diagnostics */
1104 					iflib_rx_miss ++;
1105 					iflib_rx_miss_bufs += n;
1106 				}
1107 				fl->ifl_cidx = nic_i;
1108 				kring->nr_hwtail = netmap_idx_k2n(kring, nm_i);
1109 			}
1110 			kring->nr_kflags &= ~NKR_PENDINTR;
1111 		}
1112 	}
1113 	/*
1114 	 * Second part: skip past packets that userspace has released.
1115 	 * (kring->nr_hwcur to head excluded),
1116 	 * and make the buffers available for reception.
1117 	 * As usual nm_i is the index in the netmap ring,
1118 	 * nic_i is the index in the NIC ring, and
1119 	 * nm_i == (nic_i + kring->nkr_hwofs) % ring_size
1120 	 */
1121 	/* XXX not sure how this will work with multiple free lists */
1122 	nm_i = netmap_idx_n2k(kring, kring->nr_hwcur);
1123 
1124 	return (netmap_fl_refill(rxq, kring, nm_i, false));
1125 }
1126 
1127 static void
1128 iflib_netmap_intr(struct netmap_adapter *na, int onoff)
1129 {
1130 	struct ifnet *ifp = na->ifp;
1131 	if_ctx_t ctx = ifp->if_softc;
1132 
1133 	CTX_LOCK(ctx);
1134 	if (onoff) {
1135 		IFDI_INTR_ENABLE(ctx);
1136 	} else {
1137 		IFDI_INTR_DISABLE(ctx);
1138 	}
1139 	CTX_UNLOCK(ctx);
1140 }
1141 
1142 
1143 static int
1144 iflib_netmap_attach(if_ctx_t ctx)
1145 {
1146 	struct netmap_adapter na;
1147 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1148 
1149 	bzero(&na, sizeof(na));
1150 
1151 	na.ifp = ctx->ifc_ifp;
1152 	na.na_flags = NAF_BDG_MAYSLEEP;
1153 	MPASS(ctx->ifc_softc_ctx.isc_ntxqsets);
1154 	MPASS(ctx->ifc_softc_ctx.isc_nrxqsets);
1155 
1156 	na.num_tx_desc = scctx->isc_ntxd[0];
1157 	na.num_rx_desc = scctx->isc_nrxd[0];
1158 	na.nm_txsync = iflib_netmap_txsync;
1159 	na.nm_rxsync = iflib_netmap_rxsync;
1160 	na.nm_register = iflib_netmap_register;
1161 	na.nm_intr = iflib_netmap_intr;
1162 	na.num_tx_rings = ctx->ifc_softc_ctx.isc_ntxqsets;
1163 	na.num_rx_rings = ctx->ifc_softc_ctx.isc_nrxqsets;
1164 	return (netmap_attach(&na));
1165 }
1166 
1167 static void
1168 iflib_netmap_txq_init(if_ctx_t ctx, iflib_txq_t txq)
1169 {
1170 	struct netmap_adapter *na = NA(ctx->ifc_ifp);
1171 	struct netmap_slot *slot;
1172 
1173 	slot = netmap_reset(na, NR_TX, txq->ift_id, 0);
1174 	if (slot == NULL)
1175 		return;
1176 	if (txq->ift_sds.ifsd_map == NULL)
1177 		return;
1178 
1179 	for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxd[0]; i++) {
1180 
1181 		/*
1182 		 * In netmap mode, set the map for the packet buffer.
1183 		 * NOTE: Some drivers (not this one) also need to set
1184 		 * the physical buffer address in the NIC ring.
1185 		 * netmap_idx_n2k() maps a nic index, i, into the corresponding
1186 		 * netmap slot index, si
1187 		 */
1188 		int si = netmap_idx_n2k(na->tx_rings[txq->ift_id], i);
1189 		netmap_load_map(na, txq->ift_desc_tag, txq->ift_sds.ifsd_map[i], NMB(na, slot + si));
1190 	}
1191 }
1192 
1193 static void
1194 iflib_netmap_rxq_init(if_ctx_t ctx, iflib_rxq_t rxq)
1195 {
1196 	struct netmap_adapter *na = NA(ctx->ifc_ifp);
1197 	struct netmap_kring *kring = na->rx_rings[rxq->ifr_id];
1198 	struct netmap_slot *slot;
1199 	uint32_t nm_i;
1200 
1201 	slot = netmap_reset(na, NR_RX, rxq->ifr_id, 0);
1202 	if (slot == NULL)
1203 		return;
1204 	nm_i = netmap_idx_n2k(kring, 0);
1205 	netmap_fl_refill(rxq, kring, nm_i, true);
1206 }
1207 
1208 #define iflib_netmap_detach(ifp) netmap_detach(ifp)
1209 
1210 #else
1211 #define iflib_netmap_txq_init(ctx, txq)
1212 #define iflib_netmap_rxq_init(ctx, rxq)
1213 #define iflib_netmap_detach(ifp)
1214 
1215 #define iflib_netmap_attach(ctx) (0)
1216 #define netmap_rx_irq(ifp, qid, budget) (0)
1217 #define netmap_tx_irq(ifp, qid) do {} while (0)
1218 
1219 #endif
1220 
1221 #if defined(__i386__) || defined(__amd64__)
1222 static __inline void
1223 prefetch(void *x)
1224 {
1225 	__asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x));
1226 }
1227 static __inline void
1228 prefetch2cachelines(void *x)
1229 {
1230 	__asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x));
1231 #if (CACHE_LINE_SIZE < 128)
1232 	__asm volatile("prefetcht0 %0" :: "m" (*(((unsigned long *)x)+CACHE_LINE_SIZE/(sizeof(unsigned long)))));
1233 #endif
1234 }
1235 #else
1236 #define prefetch(x)
1237 #define prefetch2cachelines(x)
1238 #endif
1239 
1240 static void
1241 iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid)
1242 {
1243 	iflib_fl_t fl;
1244 
1245 	fl = &rxq->ifr_fl[flid];
1246 	iru->iru_paddrs = fl->ifl_bus_addrs;
1247 	iru->iru_vaddrs = &fl->ifl_vm_addrs[0];
1248 	iru->iru_idxs = fl->ifl_rxd_idxs;
1249 	iru->iru_qsidx = rxq->ifr_id;
1250 	iru->iru_buf_size = fl->ifl_buf_size;
1251 	iru->iru_flidx = fl->ifl_id;
1252 }
1253 
1254 static void
1255 _iflib_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int err)
1256 {
1257 	if (err)
1258 		return;
1259 	*(bus_addr_t *) arg = segs[0].ds_addr;
1260 }
1261 
1262 int
1263 iflib_dma_alloc(if_ctx_t ctx, int size, iflib_dma_info_t dma, int mapflags)
1264 {
1265 	int err;
1266 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1267 	device_t dev = ctx->ifc_dev;
1268 
1269 	KASSERT(sctx->isc_q_align != 0, ("alignment value not initialized"));
1270 
1271 	err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
1272 				sctx->isc_q_align, 0,	/* alignment, bounds */
1273 				BUS_SPACE_MAXADDR,	/* lowaddr */
1274 				BUS_SPACE_MAXADDR,	/* highaddr */
1275 				NULL, NULL,		/* filter, filterarg */
1276 				size,			/* maxsize */
1277 				1,			/* nsegments */
1278 				size,			/* maxsegsize */
1279 				BUS_DMA_ALLOCNOW,	/* flags */
1280 				NULL,			/* lockfunc */
1281 				NULL,			/* lockarg */
1282 				&dma->idi_tag);
1283 	if (err) {
1284 		device_printf(dev,
1285 		    "%s: bus_dma_tag_create failed: %d\n",
1286 		    __func__, err);
1287 		goto fail_0;
1288 	}
1289 
1290 	err = bus_dmamem_alloc(dma->idi_tag, (void**) &dma->idi_vaddr,
1291 	    BUS_DMA_NOWAIT | BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->idi_map);
1292 	if (err) {
1293 		device_printf(dev,
1294 		    "%s: bus_dmamem_alloc(%ju) failed: %d\n",
1295 		    __func__, (uintmax_t)size, err);
1296 		goto fail_1;
1297 	}
1298 
1299 	dma->idi_paddr = IF_BAD_DMA;
1300 	err = bus_dmamap_load(dma->idi_tag, dma->idi_map, dma->idi_vaddr,
1301 	    size, _iflib_dmamap_cb, &dma->idi_paddr, mapflags | BUS_DMA_NOWAIT);
1302 	if (err || dma->idi_paddr == IF_BAD_DMA) {
1303 		device_printf(dev,
1304 		    "%s: bus_dmamap_load failed: %d\n",
1305 		    __func__, err);
1306 		goto fail_2;
1307 	}
1308 
1309 	dma->idi_size = size;
1310 	return (0);
1311 
1312 fail_2:
1313 	bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map);
1314 fail_1:
1315 	bus_dma_tag_destroy(dma->idi_tag);
1316 fail_0:
1317 	dma->idi_tag = NULL;
1318 
1319 	return (err);
1320 }
1321 
1322 int
1323 iflib_dma_alloc_multi(if_ctx_t ctx, int *sizes, iflib_dma_info_t *dmalist, int mapflags, int count)
1324 {
1325 	int i, err;
1326 	iflib_dma_info_t *dmaiter;
1327 
1328 	dmaiter = dmalist;
1329 	for (i = 0; i < count; i++, dmaiter++) {
1330 		if ((err = iflib_dma_alloc(ctx, sizes[i], *dmaiter, mapflags)) != 0)
1331 			break;
1332 	}
1333 	if (err)
1334 		iflib_dma_free_multi(dmalist, i);
1335 	return (err);
1336 }
1337 
1338 void
1339 iflib_dma_free(iflib_dma_info_t dma)
1340 {
1341 	if (dma->idi_tag == NULL)
1342 		return;
1343 	if (dma->idi_paddr != IF_BAD_DMA) {
1344 		bus_dmamap_sync(dma->idi_tag, dma->idi_map,
1345 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1346 		bus_dmamap_unload(dma->idi_tag, dma->idi_map);
1347 		dma->idi_paddr = IF_BAD_DMA;
1348 	}
1349 	if (dma->idi_vaddr != NULL) {
1350 		bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map);
1351 		dma->idi_vaddr = NULL;
1352 	}
1353 	bus_dma_tag_destroy(dma->idi_tag);
1354 	dma->idi_tag = NULL;
1355 }
1356 
1357 void
1358 iflib_dma_free_multi(iflib_dma_info_t *dmalist, int count)
1359 {
1360 	int i;
1361 	iflib_dma_info_t *dmaiter = dmalist;
1362 
1363 	for (i = 0; i < count; i++, dmaiter++)
1364 		iflib_dma_free(*dmaiter);
1365 }
1366 
1367 #ifdef EARLY_AP_STARTUP
1368 static const int iflib_started = 1;
1369 #else
1370 /*
1371  * We used to abuse the smp_started flag to decide if the queues have been
1372  * fully initialized (by late taskqgroup_adjust() calls in a SYSINIT()).
1373  * That gave bad races, since the SYSINIT() runs strictly after smp_started
1374  * is set.  Run a SYSINIT() strictly after that to just set a usable
1375  * completion flag.
1376  */
1377 
1378 static int iflib_started;
1379 
1380 static void
1381 iflib_record_started(void *arg)
1382 {
1383 	iflib_started = 1;
1384 }
1385 
1386 SYSINIT(iflib_record_started, SI_SUB_SMP + 1, SI_ORDER_FIRST,
1387 	iflib_record_started, NULL);
1388 #endif
1389 
1390 static int
1391 iflib_fast_intr(void *arg)
1392 {
1393 	iflib_filter_info_t info = arg;
1394 	struct grouptask *gtask = info->ifi_task;
1395 	if (!iflib_started)
1396 		return (FILTER_HANDLED);
1397 
1398 	DBG_COUNTER_INC(fast_intrs);
1399 	if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED)
1400 		return (FILTER_HANDLED);
1401 
1402 	GROUPTASK_ENQUEUE(gtask);
1403 	return (FILTER_HANDLED);
1404 }
1405 
1406 static int
1407 iflib_fast_intr_rxtx(void *arg)
1408 {
1409 	iflib_filter_info_t info = arg;
1410 	struct grouptask *gtask = info->ifi_task;
1411 	iflib_rxq_t rxq = (iflib_rxq_t)info->ifi_ctx;
1412 	if_ctx_t ctx;
1413 	int i, cidx;
1414 
1415 	if (!iflib_started)
1416 		return (FILTER_HANDLED);
1417 
1418 	DBG_COUNTER_INC(fast_intrs);
1419 	if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED)
1420 		return (FILTER_HANDLED);
1421 
1422 	for (i = 0; i < rxq->ifr_ntxqirq; i++) {
1423 		qidx_t txqid = rxq->ifr_txqid[i];
1424 
1425 		ctx = rxq->ifr_ctx;
1426 
1427 		if (!ctx->isc_txd_credits_update(ctx->ifc_softc, txqid, false)) {
1428 			IFDI_TX_QUEUE_INTR_ENABLE(ctx, txqid);
1429 			continue;
1430 		}
1431 		GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task);
1432 	}
1433 	if (ctx->ifc_sctx->isc_flags & IFLIB_HAS_RXCQ)
1434 		cidx = rxq->ifr_cq_cidx;
1435 	else
1436 		cidx = rxq->ifr_fl[0].ifl_cidx;
1437 	if (iflib_rxd_avail(ctx, rxq, cidx, 1))
1438 		GROUPTASK_ENQUEUE(gtask);
1439 	else
1440 		IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id);
1441 	return (FILTER_HANDLED);
1442 }
1443 
1444 
1445 static int
1446 iflib_fast_intr_ctx(void *arg)
1447 {
1448 	iflib_filter_info_t info = arg;
1449 	struct grouptask *gtask = info->ifi_task;
1450 
1451 	if (!iflib_started)
1452 		return (FILTER_HANDLED);
1453 
1454 	DBG_COUNTER_INC(fast_intrs);
1455 	if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED)
1456 		return (FILTER_HANDLED);
1457 
1458 	GROUPTASK_ENQUEUE(gtask);
1459 	return (FILTER_HANDLED);
1460 }
1461 
1462 static int
1463 _iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid,
1464 	driver_filter_t filter, driver_intr_t handler, void *arg,
1465 				 char *name)
1466 {
1467 	int rc, flags;
1468 	struct resource *res;
1469 	void *tag = NULL;
1470 	device_t dev = ctx->ifc_dev;
1471 
1472 	flags = RF_ACTIVE;
1473 	if (ctx->ifc_flags & IFC_LEGACY)
1474 		flags |= RF_SHAREABLE;
1475 	MPASS(rid < 512);
1476 	irq->ii_rid = rid;
1477 	res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &irq->ii_rid, flags);
1478 	if (res == NULL) {
1479 		device_printf(dev,
1480 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
1481 		return (ENOMEM);
1482 	}
1483 	irq->ii_res = res;
1484 	KASSERT(filter == NULL || handler == NULL, ("filter and handler can't both be non-NULL"));
1485 	rc = bus_setup_intr(dev, res, INTR_MPSAFE | INTR_TYPE_NET,
1486 						filter, handler, arg, &tag);
1487 	if (rc != 0) {
1488 		device_printf(dev,
1489 		    "failed to setup interrupt for rid %d, name %s: %d\n",
1490 					  rid, name ? name : "unknown", rc);
1491 		return (rc);
1492 	} else if (name)
1493 		bus_describe_intr(dev, res, tag, "%s", name);
1494 
1495 	irq->ii_tag = tag;
1496 	return (0);
1497 }
1498 
1499 
1500 /*********************************************************************
1501  *
1502  *  Allocate memory for tx_buffer structures. The tx_buffer stores all
1503  *  the information needed to transmit a packet on the wire. This is
1504  *  called only once at attach, setup is done every reset.
1505  *
1506  **********************************************************************/
1507 
1508 static int
1509 iflib_txsd_alloc(iflib_txq_t txq)
1510 {
1511 	if_ctx_t ctx = txq->ift_ctx;
1512 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1513 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1514 	device_t dev = ctx->ifc_dev;
1515 	int err, nsegments, ntsosegments;
1516 
1517 	nsegments = scctx->isc_tx_nsegments;
1518 	ntsosegments = scctx->isc_tx_tso_segments_max;
1519 	MPASS(scctx->isc_ntxd[0] > 0);
1520 	MPASS(scctx->isc_ntxd[txq->ift_br_offset] > 0);
1521 	MPASS(nsegments > 0);
1522 	MPASS(ntsosegments > 0);
1523 	/*
1524 	 * Setup DMA descriptor areas.
1525 	 */
1526 	if ((err = bus_dma_tag_create(bus_get_dma_tag(dev),
1527 			       1, 0,			/* alignment, bounds */
1528 			       BUS_SPACE_MAXADDR,	/* lowaddr */
1529 			       BUS_SPACE_MAXADDR,	/* highaddr */
1530 			       NULL, NULL,		/* filter, filterarg */
1531 			       sctx->isc_tx_maxsize,		/* maxsize */
1532 			       nsegments,	/* nsegments */
1533 			       sctx->isc_tx_maxsegsize,	/* maxsegsize */
1534 			       0,			/* flags */
1535 			       NULL,			/* lockfunc */
1536 			       NULL,			/* lockfuncarg */
1537 			       &txq->ift_desc_tag))) {
1538 		device_printf(dev,"Unable to allocate TX DMA tag: %d\n", err);
1539 		device_printf(dev,"maxsize: %ju nsegments: %d maxsegsize: %ju\n",
1540 		    (uintmax_t)sctx->isc_tx_maxsize, nsegments, (uintmax_t)sctx->isc_tx_maxsegsize);
1541 		goto fail;
1542 	}
1543 	if ((err = bus_dma_tag_create(bus_get_dma_tag(dev),
1544 			       1, 0,			/* alignment, bounds */
1545 			       BUS_SPACE_MAXADDR,	/* lowaddr */
1546 			       BUS_SPACE_MAXADDR,	/* highaddr */
1547 			       NULL, NULL,		/* filter, filterarg */
1548 			       scctx->isc_tx_tso_size_max,		/* maxsize */
1549 			       ntsosegments,	/* nsegments */
1550 			       scctx->isc_tx_tso_segsize_max,	/* maxsegsize */
1551 			       0,			/* flags */
1552 			       NULL,			/* lockfunc */
1553 			       NULL,			/* lockfuncarg */
1554 			       &txq->ift_tso_desc_tag))) {
1555 		device_printf(dev,"Unable to allocate TX TSO DMA tag: %d\n", err);
1556 
1557 		goto fail;
1558 	}
1559 	if (!(txq->ift_sds.ifsd_flags =
1560 	    (uint8_t *) malloc(sizeof(uint8_t) *
1561 	    scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1562 		device_printf(dev, "Unable to allocate tx_buffer memory\n");
1563 		err = ENOMEM;
1564 		goto fail;
1565 	}
1566 	if (!(txq->ift_sds.ifsd_m =
1567 	    (struct mbuf **) malloc(sizeof(struct mbuf *) *
1568 	    scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1569 		device_printf(dev, "Unable to allocate tx_buffer memory\n");
1570 		err = ENOMEM;
1571 		goto fail;
1572 	}
1573 
1574         /* Create the descriptor buffer dma maps */
1575 #if defined(ACPI_DMAR) || (! (defined(__i386__) || defined(__amd64__)))
1576 	if ((ctx->ifc_flags & IFC_DMAR) == 0)
1577 		return (0);
1578 
1579 	if (!(txq->ift_sds.ifsd_map =
1580 	    (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1581 		device_printf(dev, "Unable to allocate tx_buffer map memory\n");
1582 		err = ENOMEM;
1583 		goto fail;
1584 	}
1585 
1586 	for (int i = 0; i < scctx->isc_ntxd[txq->ift_br_offset]; i++) {
1587 		err = bus_dmamap_create(txq->ift_desc_tag, 0, &txq->ift_sds.ifsd_map[i]);
1588 		if (err != 0) {
1589 			device_printf(dev, "Unable to create TX DMA map\n");
1590 			goto fail;
1591 		}
1592 	}
1593 #endif
1594 	return (0);
1595 fail:
1596 	/* We free all, it handles case where we are in the middle */
1597 	iflib_tx_structures_free(ctx);
1598 	return (err);
1599 }
1600 
1601 static void
1602 iflib_txsd_destroy(if_ctx_t ctx, iflib_txq_t txq, int i)
1603 {
1604 	bus_dmamap_t map;
1605 
1606 	map = NULL;
1607 	if (txq->ift_sds.ifsd_map != NULL)
1608 		map = txq->ift_sds.ifsd_map[i];
1609 	if (map != NULL) {
1610 		bus_dmamap_unload(txq->ift_desc_tag, map);
1611 		bus_dmamap_destroy(txq->ift_desc_tag, map);
1612 		txq->ift_sds.ifsd_map[i] = NULL;
1613 	}
1614 }
1615 
1616 static void
1617 iflib_txq_destroy(iflib_txq_t txq)
1618 {
1619 	if_ctx_t ctx = txq->ift_ctx;
1620 
1621 	for (int i = 0; i < txq->ift_size; i++)
1622 		iflib_txsd_destroy(ctx, txq, i);
1623 	if (txq->ift_sds.ifsd_map != NULL) {
1624 		free(txq->ift_sds.ifsd_map, M_IFLIB);
1625 		txq->ift_sds.ifsd_map = NULL;
1626 	}
1627 	if (txq->ift_sds.ifsd_m != NULL) {
1628 		free(txq->ift_sds.ifsd_m, M_IFLIB);
1629 		txq->ift_sds.ifsd_m = NULL;
1630 	}
1631 	if (txq->ift_sds.ifsd_flags != NULL) {
1632 		free(txq->ift_sds.ifsd_flags, M_IFLIB);
1633 		txq->ift_sds.ifsd_flags = NULL;
1634 	}
1635 	if (txq->ift_desc_tag != NULL) {
1636 		bus_dma_tag_destroy(txq->ift_desc_tag);
1637 		txq->ift_desc_tag = NULL;
1638 	}
1639 	if (txq->ift_tso_desc_tag != NULL) {
1640 		bus_dma_tag_destroy(txq->ift_tso_desc_tag);
1641 		txq->ift_tso_desc_tag = NULL;
1642 	}
1643 }
1644 
1645 static void
1646 iflib_txsd_free(if_ctx_t ctx, iflib_txq_t txq, int i)
1647 {
1648 	struct mbuf **mp;
1649 
1650 	mp = &txq->ift_sds.ifsd_m[i];
1651 	if (*mp == NULL)
1652 		return;
1653 
1654 	if (txq->ift_sds.ifsd_map != NULL) {
1655 		bus_dmamap_sync(txq->ift_desc_tag,
1656 				txq->ift_sds.ifsd_map[i],
1657 				BUS_DMASYNC_POSTWRITE);
1658 		bus_dmamap_unload(txq->ift_desc_tag,
1659 				  txq->ift_sds.ifsd_map[i]);
1660 	}
1661 	m_free(*mp);
1662 	DBG_COUNTER_INC(tx_frees);
1663 	*mp = NULL;
1664 }
1665 
1666 static int
1667 iflib_txq_setup(iflib_txq_t txq)
1668 {
1669 	if_ctx_t ctx = txq->ift_ctx;
1670 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1671 	iflib_dma_info_t di;
1672 	int i;
1673 
1674 	/* Set number of descriptors available */
1675 	txq->ift_qstatus = IFLIB_QUEUE_IDLE;
1676 	/* XXX make configurable */
1677 	txq->ift_update_freq = IFLIB_DEFAULT_TX_UPDATE_FREQ;
1678 
1679 	/* Reset indices */
1680 	txq->ift_cidx_processed = 0;
1681 	txq->ift_pidx = txq->ift_cidx = txq->ift_npending = 0;
1682 	txq->ift_size = scctx->isc_ntxd[txq->ift_br_offset];
1683 
1684 	for (i = 0, di = txq->ift_ifdi; i < ctx->ifc_nhwtxqs; i++, di++)
1685 		bzero((void *)di->idi_vaddr, di->idi_size);
1686 
1687 	IFDI_TXQ_SETUP(ctx, txq->ift_id);
1688 	for (i = 0, di = txq->ift_ifdi; i < ctx->ifc_nhwtxqs; i++, di++)
1689 		bus_dmamap_sync(di->idi_tag, di->idi_map,
1690 						BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1691 	return (0);
1692 }
1693 
1694 /*********************************************************************
1695  *
1696  *  Allocate memory for rx_buffer structures. Since we use one
1697  *  rx_buffer per received packet, the maximum number of rx_buffer's
1698  *  that we'll need is equal to the number of receive descriptors
1699  *  that we've allocated.
1700  *
1701  **********************************************************************/
1702 static int
1703 iflib_rxsd_alloc(iflib_rxq_t rxq)
1704 {
1705 	if_ctx_t ctx = rxq->ifr_ctx;
1706 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1707 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1708 	device_t dev = ctx->ifc_dev;
1709 	iflib_fl_t fl;
1710 	int			err;
1711 
1712 	MPASS(scctx->isc_nrxd[0] > 0);
1713 	MPASS(scctx->isc_nrxd[rxq->ifr_fl_offset] > 0);
1714 
1715 	fl = rxq->ifr_fl;
1716 	for (int i = 0; i <  rxq->ifr_nfl; i++, fl++) {
1717 		fl->ifl_size = scctx->isc_nrxd[rxq->ifr_fl_offset]; /* this isn't necessarily the same */
1718 		err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
1719 					 1, 0,			/* alignment, bounds */
1720 					 BUS_SPACE_MAXADDR,	/* lowaddr */
1721 					 BUS_SPACE_MAXADDR,	/* highaddr */
1722 					 NULL, NULL,		/* filter, filterarg */
1723 					 sctx->isc_rx_maxsize,	/* maxsize */
1724 					 sctx->isc_rx_nsegments,	/* nsegments */
1725 					 sctx->isc_rx_maxsegsize,	/* maxsegsize */
1726 					 0,			/* flags */
1727 					 NULL,			/* lockfunc */
1728 					 NULL,			/* lockarg */
1729 					 &fl->ifl_desc_tag);
1730 		if (err) {
1731 			device_printf(dev, "%s: bus_dma_tag_create failed %d\n",
1732 				__func__, err);
1733 			goto fail;
1734 		}
1735 		if (!(fl->ifl_sds.ifsd_flags =
1736 		      (uint8_t *) malloc(sizeof(uint8_t) *
1737 					 scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1738 			device_printf(dev, "Unable to allocate tx_buffer memory\n");
1739 			err = ENOMEM;
1740 			goto fail;
1741 		}
1742 		if (!(fl->ifl_sds.ifsd_m =
1743 		      (struct mbuf **) malloc(sizeof(struct mbuf *) *
1744 					      scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1745 			device_printf(dev, "Unable to allocate tx_buffer memory\n");
1746 			err = ENOMEM;
1747 			goto fail;
1748 		}
1749 		if (!(fl->ifl_sds.ifsd_cl =
1750 		      (caddr_t *) malloc(sizeof(caddr_t) *
1751 					      scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1752 			device_printf(dev, "Unable to allocate tx_buffer memory\n");
1753 			err = ENOMEM;
1754 			goto fail;
1755 		}
1756 
1757 		/* Create the descriptor buffer dma maps */
1758 #if defined(ACPI_DMAR) || (! (defined(__i386__) || defined(__amd64__)))
1759 		if ((ctx->ifc_flags & IFC_DMAR) == 0)
1760 			continue;
1761 
1762 		if (!(fl->ifl_sds.ifsd_map =
1763 		      (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1764 			device_printf(dev, "Unable to allocate tx_buffer map memory\n");
1765 			err = ENOMEM;
1766 			goto fail;
1767 		}
1768 
1769 		for (int i = 0; i < scctx->isc_nrxd[rxq->ifr_fl_offset]; i++) {
1770 			err = bus_dmamap_create(fl->ifl_desc_tag, 0, &fl->ifl_sds.ifsd_map[i]);
1771 			if (err != 0) {
1772 				device_printf(dev, "Unable to create RX buffer DMA map\n");
1773 				goto fail;
1774 			}
1775 		}
1776 #endif
1777 	}
1778 	return (0);
1779 
1780 fail:
1781 	iflib_rx_structures_free(ctx);
1782 	return (err);
1783 }
1784 
1785 
1786 /*
1787  * Internal service routines
1788  */
1789 
1790 struct rxq_refill_cb_arg {
1791 	int               error;
1792 	bus_dma_segment_t seg;
1793 	int               nseg;
1794 };
1795 
1796 static void
1797 _rxq_refill_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1798 {
1799 	struct rxq_refill_cb_arg *cb_arg = arg;
1800 
1801 	cb_arg->error = error;
1802 	cb_arg->seg = segs[0];
1803 	cb_arg->nseg = nseg;
1804 }
1805 
1806 
1807 #ifdef ACPI_DMAR
1808 #define IS_DMAR(ctx) (ctx->ifc_flags & IFC_DMAR)
1809 #else
1810 #define IS_DMAR(ctx) (0)
1811 #endif
1812 
1813 /**
1814  *	rxq_refill - refill an rxq  free-buffer list
1815  *	@ctx: the iflib context
1816  *	@rxq: the free-list to refill
1817  *	@n: the number of new buffers to allocate
1818  *
1819  *	(Re)populate an rxq free-buffer list with up to @n new packet buffers.
1820  *	The caller must assure that @n does not exceed the queue's capacity.
1821  */
1822 static void
1823 _iflib_fl_refill(if_ctx_t ctx, iflib_fl_t fl, int count)
1824 {
1825 	struct mbuf *m;
1826 	int idx, frag_idx = fl->ifl_fragidx;
1827         int pidx = fl->ifl_pidx;
1828 	caddr_t cl, *sd_cl;
1829 	struct mbuf **sd_m;
1830 	uint8_t *sd_flags;
1831 	struct if_rxd_update iru;
1832 	bus_dmamap_t *sd_map;
1833 	int n, i = 0;
1834 	uint64_t bus_addr;
1835 	int err;
1836 	qidx_t credits;
1837 
1838 	sd_m = fl->ifl_sds.ifsd_m;
1839 	sd_map = fl->ifl_sds.ifsd_map;
1840 	sd_cl = fl->ifl_sds.ifsd_cl;
1841 	sd_flags = fl->ifl_sds.ifsd_flags;
1842 	idx = pidx;
1843 	credits = fl->ifl_credits;
1844 
1845 	n  = count;
1846 	MPASS(n > 0);
1847 	MPASS(credits + n <= fl->ifl_size);
1848 
1849 	if (pidx < fl->ifl_cidx)
1850 		MPASS(pidx + n <= fl->ifl_cidx);
1851 	if (pidx == fl->ifl_cidx && (credits < fl->ifl_size))
1852 		MPASS(fl->ifl_gen == 0);
1853 	if (pidx > fl->ifl_cidx)
1854 		MPASS(n <= fl->ifl_size - pidx + fl->ifl_cidx);
1855 
1856 	DBG_COUNTER_INC(fl_refills);
1857 	if (n > 8)
1858 		DBG_COUNTER_INC(fl_refills_large);
1859 	iru_init(&iru, fl->ifl_rxq, fl->ifl_id);
1860 	while (n--) {
1861 		/*
1862 		 * We allocate an uninitialized mbuf + cluster, mbuf is
1863 		 * initialized after rx.
1864 		 *
1865 		 * If the cluster is still set then we know a minimum sized packet was received
1866 		 */
1867 		bit_ffc_at(fl->ifl_rx_bitmap, frag_idx, fl->ifl_size,  &frag_idx);
1868 		if ((frag_idx < 0) || (frag_idx >= fl->ifl_size))
1869                 	bit_ffc(fl->ifl_rx_bitmap, fl->ifl_size, &frag_idx);
1870 		if ((cl = sd_cl[frag_idx]) == NULL) {
1871                        if ((cl = sd_cl[frag_idx] = m_cljget(NULL, M_NOWAIT, fl->ifl_buf_size)) == NULL)
1872 				break;
1873 #if MEMORY_LOGGING
1874 			fl->ifl_cl_enqueued++;
1875 #endif
1876 		}
1877 		if ((m = m_gethdr(M_NOWAIT, MT_NOINIT)) == NULL) {
1878 			break;
1879 		}
1880 #if MEMORY_LOGGING
1881 		fl->ifl_m_enqueued++;
1882 #endif
1883 
1884 		DBG_COUNTER_INC(rx_allocs);
1885 #if defined(__i386__) || defined(__amd64__)
1886 		if (!IS_DMAR(ctx)) {
1887 			bus_addr = pmap_kextract((vm_offset_t)cl);
1888 		} else
1889 #endif
1890 		{
1891 			struct rxq_refill_cb_arg cb_arg;
1892 			iflib_rxq_t q;
1893 
1894 			cb_arg.error = 0;
1895 			q = fl->ifl_rxq;
1896 			MPASS(sd_map != NULL);
1897 			MPASS(sd_map[frag_idx] != NULL);
1898 			err = bus_dmamap_load(fl->ifl_desc_tag, sd_map[frag_idx],
1899 		         cl, fl->ifl_buf_size, _rxq_refill_cb, &cb_arg, 0);
1900 			bus_dmamap_sync(fl->ifl_desc_tag, sd_map[frag_idx],
1901 					BUS_DMASYNC_PREREAD);
1902 
1903 			if (err != 0 || cb_arg.error) {
1904 				/*
1905 				 * !zone_pack ?
1906 				 */
1907 				if (fl->ifl_zone == zone_pack)
1908 					uma_zfree(fl->ifl_zone, cl);
1909 				m_free(m);
1910 				n = 0;
1911 				goto done;
1912 			}
1913 			bus_addr = cb_arg.seg.ds_addr;
1914 		}
1915                 bit_set(fl->ifl_rx_bitmap, frag_idx);
1916 		sd_flags[frag_idx] |= RX_SW_DESC_INUSE;
1917 
1918 		MPASS(sd_m[frag_idx] == NULL);
1919 		sd_cl[frag_idx] = cl;
1920 		sd_m[frag_idx] = m;
1921 		fl->ifl_rxd_idxs[i] = frag_idx;
1922 		fl->ifl_bus_addrs[i] = bus_addr;
1923 		fl->ifl_vm_addrs[i] = cl;
1924 		credits++;
1925 		i++;
1926 		MPASS(credits <= fl->ifl_size);
1927 		if (++idx == fl->ifl_size) {
1928 			fl->ifl_gen = 1;
1929 			idx = 0;
1930 		}
1931 		if (n == 0 || i == IFLIB_MAX_RX_REFRESH) {
1932 			iru.iru_pidx = pidx;
1933 			iru.iru_count = i;
1934 			ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
1935 			i = 0;
1936 			pidx = idx;
1937 			fl->ifl_pidx = idx;
1938 			fl->ifl_credits = credits;
1939 		}
1940 
1941 	}
1942 done:
1943 	if (i) {
1944 		iru.iru_pidx = pidx;
1945 		iru.iru_count = i;
1946 		ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
1947 		fl->ifl_pidx = idx;
1948 		fl->ifl_credits = credits;
1949 	}
1950 	DBG_COUNTER_INC(rxd_flush);
1951 	if (fl->ifl_pidx == 0)
1952 		pidx = fl->ifl_size - 1;
1953 	else
1954 		pidx = fl->ifl_pidx - 1;
1955 
1956 	if (sd_map)
1957 		bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
1958 				BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1959 	ctx->isc_rxd_flush(ctx->ifc_softc, fl->ifl_rxq->ifr_id, fl->ifl_id, pidx);
1960 	fl->ifl_fragidx = frag_idx;
1961 }
1962 
1963 static __inline void
1964 __iflib_fl_refill_lt(if_ctx_t ctx, iflib_fl_t fl, int max)
1965 {
1966 	/* we avoid allowing pidx to catch up with cidx as it confuses ixl */
1967 	int32_t reclaimable = fl->ifl_size - fl->ifl_credits - 1;
1968 #ifdef INVARIANTS
1969 	int32_t delta = fl->ifl_size - get_inuse(fl->ifl_size, fl->ifl_cidx, fl->ifl_pidx, fl->ifl_gen) - 1;
1970 #endif
1971 
1972 	MPASS(fl->ifl_credits <= fl->ifl_size);
1973 	MPASS(reclaimable == delta);
1974 
1975 	if (reclaimable > 0)
1976 		_iflib_fl_refill(ctx, fl, min(max, reclaimable));
1977 }
1978 
1979 static void
1980 iflib_fl_bufs_free(iflib_fl_t fl)
1981 {
1982 	iflib_dma_info_t idi = fl->ifl_ifdi;
1983 	uint32_t i;
1984 
1985 	for (i = 0; i < fl->ifl_size; i++) {
1986 		struct mbuf **sd_m = &fl->ifl_sds.ifsd_m[i];
1987 		uint8_t *sd_flags = &fl->ifl_sds.ifsd_flags[i];
1988 		caddr_t *sd_cl = &fl->ifl_sds.ifsd_cl[i];
1989 
1990 		if (*sd_flags & RX_SW_DESC_INUSE) {
1991 			if (fl->ifl_sds.ifsd_map != NULL) {
1992 				bus_dmamap_t sd_map = fl->ifl_sds.ifsd_map[i];
1993 				bus_dmamap_unload(fl->ifl_desc_tag, sd_map);
1994 				if (fl->ifl_rxq->ifr_ctx->ifc_in_detach)
1995 					bus_dmamap_destroy(fl->ifl_desc_tag, sd_map);
1996 			}
1997 			if (*sd_m != NULL) {
1998 				m_init(*sd_m, M_NOWAIT, MT_DATA, 0);
1999 				uma_zfree(zone_mbuf, *sd_m);
2000 			}
2001 			if (*sd_cl != NULL)
2002 				uma_zfree(fl->ifl_zone, *sd_cl);
2003 			*sd_flags = 0;
2004 		} else {
2005 			MPASS(*sd_cl == NULL);
2006 			MPASS(*sd_m == NULL);
2007 		}
2008 #if MEMORY_LOGGING
2009 		fl->ifl_m_dequeued++;
2010 		fl->ifl_cl_dequeued++;
2011 #endif
2012 		*sd_cl = NULL;
2013 		*sd_m = NULL;
2014 	}
2015 #ifdef INVARIANTS
2016 	for (i = 0; i < fl->ifl_size; i++) {
2017 		MPASS(fl->ifl_sds.ifsd_flags[i] == 0);
2018 		MPASS(fl->ifl_sds.ifsd_cl[i] == NULL);
2019 		MPASS(fl->ifl_sds.ifsd_m[i] == NULL);
2020 	}
2021 #endif
2022 	/*
2023 	 * Reset free list values
2024 	 */
2025 	fl->ifl_credits = fl->ifl_cidx = fl->ifl_pidx = fl->ifl_gen = fl->ifl_fragidx = 0;
2026 	bzero(idi->idi_vaddr, idi->idi_size);
2027 }
2028 
2029 /*********************************************************************
2030  *
2031  *  Initialize a receive ring and its buffers.
2032  *
2033  **********************************************************************/
2034 static int
2035 iflib_fl_setup(iflib_fl_t fl)
2036 {
2037 	iflib_rxq_t rxq = fl->ifl_rxq;
2038 	if_ctx_t ctx = rxq->ifr_ctx;
2039 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2040 
2041 	bit_nclear(fl->ifl_rx_bitmap, 0, fl->ifl_size - 1);
2042 	/*
2043 	** Free current RX buffer structs and their mbufs
2044 	*/
2045 	iflib_fl_bufs_free(fl);
2046 	/* Now replenish the mbufs */
2047 	MPASS(fl->ifl_credits == 0);
2048 	/*
2049 	 * XXX don't set the max_frame_size to larger
2050 	 * than the hardware can handle
2051 	 */
2052 	if (sctx->isc_max_frame_size <= 2048)
2053 		fl->ifl_buf_size = MCLBYTES;
2054 #ifndef CONTIGMALLOC_WORKS
2055 	else
2056 		fl->ifl_buf_size = MJUMPAGESIZE;
2057 #else
2058 	else if (sctx->isc_max_frame_size <= 4096)
2059 		fl->ifl_buf_size = MJUMPAGESIZE;
2060 	else if (sctx->isc_max_frame_size <= 9216)
2061 		fl->ifl_buf_size = MJUM9BYTES;
2062 	else
2063 		fl->ifl_buf_size = MJUM16BYTES;
2064 #endif
2065 	if (fl->ifl_buf_size > ctx->ifc_max_fl_buf_size)
2066 		ctx->ifc_max_fl_buf_size = fl->ifl_buf_size;
2067 	fl->ifl_cltype = m_gettype(fl->ifl_buf_size);
2068 	fl->ifl_zone = m_getzone(fl->ifl_buf_size);
2069 
2070 
2071 	/* avoid pre-allocating zillions of clusters to an idle card
2072 	 * potentially speeding up attach
2073 	 */
2074 	_iflib_fl_refill(ctx, fl, min(128, fl->ifl_size));
2075 	MPASS(min(128, fl->ifl_size) == fl->ifl_credits);
2076 	if (min(128, fl->ifl_size) != fl->ifl_credits)
2077 		return (ENOBUFS);
2078 	/*
2079 	 * handle failure
2080 	 */
2081 	MPASS(rxq != NULL);
2082 	MPASS(fl->ifl_ifdi != NULL);
2083 	bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
2084 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2085 	return (0);
2086 }
2087 
2088 /*********************************************************************
2089  *
2090  *  Free receive ring data structures
2091  *
2092  **********************************************************************/
2093 static void
2094 iflib_rx_sds_free(iflib_rxq_t rxq)
2095 {
2096 	iflib_fl_t fl;
2097 	int i;
2098 
2099 	if (rxq->ifr_fl != NULL) {
2100 		for (i = 0; i < rxq->ifr_nfl; i++) {
2101 			fl = &rxq->ifr_fl[i];
2102 			if (fl->ifl_desc_tag != NULL) {
2103 				bus_dma_tag_destroy(fl->ifl_desc_tag);
2104 				fl->ifl_desc_tag = NULL;
2105 			}
2106 			free(fl->ifl_sds.ifsd_m, M_IFLIB);
2107 			free(fl->ifl_sds.ifsd_cl, M_IFLIB);
2108 			/* XXX destroy maps first */
2109 			free(fl->ifl_sds.ifsd_map, M_IFLIB);
2110 			fl->ifl_sds.ifsd_m = NULL;
2111 			fl->ifl_sds.ifsd_cl = NULL;
2112 			fl->ifl_sds.ifsd_map = NULL;
2113 		}
2114 		free(rxq->ifr_fl, M_IFLIB);
2115 		rxq->ifr_fl = NULL;
2116 		rxq->ifr_cq_gen = rxq->ifr_cq_cidx = rxq->ifr_cq_pidx = 0;
2117 	}
2118 }
2119 
2120 /*
2121  * MI independent logic
2122  *
2123  */
2124 static void
2125 iflib_timer(void *arg)
2126 {
2127 	iflib_txq_t txq = arg;
2128 	if_ctx_t ctx = txq->ift_ctx;
2129 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2130 
2131 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))
2132 		return;
2133 	/*
2134 	** Check on the state of the TX queue(s), this
2135 	** can be done without the lock because its RO
2136 	** and the HUNG state will be static if set.
2137 	*/
2138 	IFDI_TIMER(ctx, txq->ift_id);
2139 	if ((txq->ift_qstatus == IFLIB_QUEUE_HUNG) &&
2140 	    ((txq->ift_cleaned_prev == txq->ift_cleaned) ||
2141 	     (sctx->isc_pause_frames == 0)))
2142 		goto hung;
2143 
2144 	if (ifmp_ring_is_stalled(txq->ift_br))
2145 		txq->ift_qstatus = IFLIB_QUEUE_HUNG;
2146 	txq->ift_cleaned_prev = txq->ift_cleaned;
2147 	/* handle any laggards */
2148 	if (txq->ift_db_pending)
2149 		GROUPTASK_ENQUEUE(&txq->ift_task);
2150 
2151 	sctx->isc_pause_frames = 0;
2152 	if (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)
2153 		callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu);
2154 	return;
2155  hung:
2156 	device_printf(ctx->ifc_dev,  "TX(%d) desc avail = %d, pidx = %d\n",
2157 				  txq->ift_id, TXQ_AVAIL(txq), txq->ift_pidx);
2158 	STATE_LOCK(ctx);
2159 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2160 	ctx->ifc_flags |= (IFC_DO_WATCHDOG|IFC_DO_RESET);
2161 	iflib_admin_intr_deferred(ctx);
2162 	STATE_UNLOCK(ctx);
2163 }
2164 
2165 static void
2166 iflib_init_locked(if_ctx_t ctx)
2167 {
2168 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2169 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2170 	if_t ifp = ctx->ifc_ifp;
2171 	iflib_fl_t fl;
2172 	iflib_txq_t txq;
2173 	iflib_rxq_t rxq;
2174 	int i, j, tx_ip_csum_flags, tx_ip6_csum_flags;
2175 
2176 
2177 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2178 	IFDI_INTR_DISABLE(ctx);
2179 
2180 	tx_ip_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP);
2181 	tx_ip6_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_UDP | CSUM_IP6_SCTP);
2182 	/* Set hardware offload abilities */
2183 	if_clearhwassist(ifp);
2184 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
2185 		if_sethwassistbits(ifp, tx_ip_csum_flags, 0);
2186 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
2187 		if_sethwassistbits(ifp,  tx_ip6_csum_flags, 0);
2188 	if (if_getcapenable(ifp) & IFCAP_TSO4)
2189 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
2190 	if (if_getcapenable(ifp) & IFCAP_TSO6)
2191 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
2192 
2193 	for (i = 0, txq = ctx->ifc_txqs; i < sctx->isc_ntxqsets; i++, txq++) {
2194 		CALLOUT_LOCK(txq);
2195 		callout_stop(&txq->ift_timer);
2196 		CALLOUT_UNLOCK(txq);
2197 		iflib_netmap_txq_init(ctx, txq);
2198 	}
2199 #ifdef INVARIANTS
2200 	i = if_getdrvflags(ifp);
2201 #endif
2202 	IFDI_INIT(ctx);
2203 	MPASS(if_getdrvflags(ifp) == i);
2204 	for (i = 0, rxq = ctx->ifc_rxqs; i < sctx->isc_nrxqsets; i++, rxq++) {
2205 		/* XXX this should really be done on a per-queue basis */
2206 		if (if_getcapenable(ifp) & IFCAP_NETMAP) {
2207 			MPASS(rxq->ifr_id == i);
2208 			iflib_netmap_rxq_init(ctx, rxq);
2209 			continue;
2210 		}
2211 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) {
2212 			if (iflib_fl_setup(fl)) {
2213 				device_printf(ctx->ifc_dev, "freelist setup failed - check cluster settings\n");
2214 				goto done;
2215 			}
2216 		}
2217 	}
2218 	done:
2219 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
2220 	IFDI_INTR_ENABLE(ctx);
2221 	txq = ctx->ifc_txqs;
2222 	for (i = 0; i < sctx->isc_ntxqsets; i++, txq++)
2223 		callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq,
2224 			txq->ift_timer.c_cpu);
2225 }
2226 
2227 static int
2228 iflib_media_change(if_t ifp)
2229 {
2230 	if_ctx_t ctx = if_getsoftc(ifp);
2231 	int err;
2232 
2233 	CTX_LOCK(ctx);
2234 	if ((err = IFDI_MEDIA_CHANGE(ctx)) == 0)
2235 		iflib_init_locked(ctx);
2236 	CTX_UNLOCK(ctx);
2237 	return (err);
2238 }
2239 
2240 static void
2241 iflib_media_status(if_t ifp, struct ifmediareq *ifmr)
2242 {
2243 	if_ctx_t ctx = if_getsoftc(ifp);
2244 
2245 	CTX_LOCK(ctx);
2246 	IFDI_UPDATE_ADMIN_STATUS(ctx);
2247 	IFDI_MEDIA_STATUS(ctx, ifmr);
2248 	CTX_UNLOCK(ctx);
2249 }
2250 
2251 static void
2252 iflib_stop(if_ctx_t ctx)
2253 {
2254 	iflib_txq_t txq = ctx->ifc_txqs;
2255 	iflib_rxq_t rxq = ctx->ifc_rxqs;
2256 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2257 	iflib_dma_info_t di;
2258 	iflib_fl_t fl;
2259 	int i, j;
2260 
2261 	/* Tell the stack that the interface is no longer active */
2262 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2263 
2264 	IFDI_INTR_DISABLE(ctx);
2265 	DELAY(1000);
2266 	IFDI_STOP(ctx);
2267 	DELAY(1000);
2268 
2269 	iflib_debug_reset();
2270 	/* Wait for current tx queue users to exit to disarm watchdog timer. */
2271 	for (i = 0; i < scctx->isc_ntxqsets; i++, txq++) {
2272 		/* make sure all transmitters have completed before proceeding XXX */
2273 
2274 		CALLOUT_LOCK(txq);
2275 		callout_stop(&txq->ift_timer);
2276 		CALLOUT_UNLOCK(txq);
2277 
2278 		/* clean any enqueued buffers */
2279 		iflib_ifmp_purge(txq);
2280 		/* Free any existing tx buffers. */
2281 		for (j = 0; j < txq->ift_size; j++) {
2282 			iflib_txsd_free(ctx, txq, j);
2283 		}
2284 		txq->ift_processed = txq->ift_cleaned = txq->ift_cidx_processed = 0;
2285 		txq->ift_in_use = txq->ift_gen = txq->ift_cidx = txq->ift_pidx = txq->ift_no_desc_avail = 0;
2286 		txq->ift_closed = txq->ift_mbuf_defrag = txq->ift_mbuf_defrag_failed = 0;
2287 		txq->ift_no_tx_dma_setup = txq->ift_txd_encap_efbig = txq->ift_map_failed = 0;
2288 		txq->ift_pullups = 0;
2289 		ifmp_ring_reset_stats(txq->ift_br);
2290 		for (j = 0, di = txq->ift_ifdi; j < ctx->ifc_nhwtxqs; j++, di++)
2291 			bzero((void *)di->idi_vaddr, di->idi_size);
2292 	}
2293 	for (i = 0; i < scctx->isc_nrxqsets; i++, rxq++) {
2294 		/* make sure all transmitters have completed before proceeding XXX */
2295 
2296 		for (j = 0, di = rxq->ifr_ifdi; j < rxq->ifr_nfl; j++, di++)
2297 			bzero((void *)di->idi_vaddr, di->idi_size);
2298 		/* also resets the free lists pidx/cidx */
2299 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++)
2300 			iflib_fl_bufs_free(fl);
2301 	}
2302 }
2303 
2304 static inline caddr_t
2305 calc_next_rxd(iflib_fl_t fl, int cidx)
2306 {
2307 	qidx_t size;
2308 	int nrxd;
2309 	caddr_t start, end, cur, next;
2310 
2311 	nrxd = fl->ifl_size;
2312 	size = fl->ifl_rxd_size;
2313 	start = fl->ifl_ifdi->idi_vaddr;
2314 
2315 	if (__predict_false(size == 0))
2316 		return (start);
2317 	cur = start + size*cidx;
2318 	end = start + size*nrxd;
2319 	next = CACHE_PTR_NEXT(cur);
2320 	return (next < end ? next : start);
2321 }
2322 
2323 static inline void
2324 prefetch_pkts(iflib_fl_t fl, int cidx)
2325 {
2326 	int nextptr;
2327 	int nrxd = fl->ifl_size;
2328 	caddr_t next_rxd;
2329 
2330 
2331 	nextptr = (cidx + CACHE_PTR_INCREMENT) & (nrxd-1);
2332 	prefetch(&fl->ifl_sds.ifsd_m[nextptr]);
2333 	prefetch(&fl->ifl_sds.ifsd_cl[nextptr]);
2334 	next_rxd = calc_next_rxd(fl, cidx);
2335 	prefetch(next_rxd);
2336 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 1) & (nrxd-1)]);
2337 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 2) & (nrxd-1)]);
2338 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 3) & (nrxd-1)]);
2339 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 4) & (nrxd-1)]);
2340 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 1) & (nrxd-1)]);
2341 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 2) & (nrxd-1)]);
2342 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 3) & (nrxd-1)]);
2343 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 4) & (nrxd-1)]);
2344 }
2345 
2346 static void
2347 rxd_frag_to_sd(iflib_rxq_t rxq, if_rxd_frag_t irf, int unload, if_rxsd_t sd)
2348 {
2349 	int flid, cidx;
2350 	bus_dmamap_t map;
2351 	iflib_fl_t fl;
2352 	iflib_dma_info_t di;
2353 	int next;
2354 
2355 	map = NULL;
2356 	flid = irf->irf_flid;
2357 	cidx = irf->irf_idx;
2358 	fl = &rxq->ifr_fl[flid];
2359 	sd->ifsd_fl = fl;
2360 	sd->ifsd_cidx = cidx;
2361 	sd->ifsd_m = &fl->ifl_sds.ifsd_m[cidx];
2362 	sd->ifsd_cl = &fl->ifl_sds.ifsd_cl[cidx];
2363 	fl->ifl_credits--;
2364 #if MEMORY_LOGGING
2365 	fl->ifl_m_dequeued++;
2366 #endif
2367 	if (rxq->ifr_ctx->ifc_flags & IFC_PREFETCH)
2368 		prefetch_pkts(fl, cidx);
2369 	if (fl->ifl_sds.ifsd_map != NULL) {
2370 		next = (cidx + CACHE_PTR_INCREMENT) & (fl->ifl_size-1);
2371 		prefetch(&fl->ifl_sds.ifsd_map[next]);
2372 		map = fl->ifl_sds.ifsd_map[cidx];
2373 		di = fl->ifl_ifdi;
2374 		next = (cidx + CACHE_LINE_SIZE) & (fl->ifl_size-1);
2375 		prefetch(&fl->ifl_sds.ifsd_flags[next]);
2376 		bus_dmamap_sync(di->idi_tag, di->idi_map,
2377 				BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2378 
2379 	/* not valid assert if bxe really does SGE from non-contiguous elements */
2380 		MPASS(fl->ifl_cidx == cidx);
2381 		if (unload)
2382 			bus_dmamap_unload(fl->ifl_desc_tag, map);
2383 	}
2384 	fl->ifl_cidx = (fl->ifl_cidx + 1) & (fl->ifl_size-1);
2385 	if (__predict_false(fl->ifl_cidx == 0))
2386 		fl->ifl_gen = 0;
2387 	if (map != NULL)
2388 		bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
2389 			BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2390         bit_clear(fl->ifl_rx_bitmap, cidx);
2391 }
2392 
2393 static struct mbuf *
2394 assemble_segments(iflib_rxq_t rxq, if_rxd_info_t ri, if_rxsd_t sd)
2395 {
2396 	int i, padlen , flags;
2397 	struct mbuf *m, *mh, *mt;
2398 	caddr_t cl;
2399 
2400 	i = 0;
2401 	mh = NULL;
2402 	do {
2403 		rxd_frag_to_sd(rxq, &ri->iri_frags[i], TRUE, sd);
2404 
2405 		MPASS(*sd->ifsd_cl != NULL);
2406 		MPASS(*sd->ifsd_m != NULL);
2407 
2408 		/* Don't include zero-length frags */
2409 		if (ri->iri_frags[i].irf_len == 0) {
2410 			/* XXX we can save the cluster here, but not the mbuf */
2411 			m_init(*sd->ifsd_m, M_NOWAIT, MT_DATA, 0);
2412 			m_free(*sd->ifsd_m);
2413 			*sd->ifsd_m = NULL;
2414 			continue;
2415 		}
2416 		m = *sd->ifsd_m;
2417 		*sd->ifsd_m = NULL;
2418 		if (mh == NULL) {
2419 			flags = M_PKTHDR|M_EXT;
2420 			mh = mt = m;
2421 			padlen = ri->iri_pad;
2422 		} else {
2423 			flags = M_EXT;
2424 			mt->m_next = m;
2425 			mt = m;
2426 			/* assuming padding is only on the first fragment */
2427 			padlen = 0;
2428 		}
2429 		cl = *sd->ifsd_cl;
2430 		*sd->ifsd_cl = NULL;
2431 
2432 		/* Can these two be made one ? */
2433 		m_init(m, M_NOWAIT, MT_DATA, flags);
2434 		m_cljset(m, cl, sd->ifsd_fl->ifl_cltype);
2435 		/*
2436 		 * These must follow m_init and m_cljset
2437 		 */
2438 		m->m_data += padlen;
2439 		ri->iri_len -= padlen;
2440 		m->m_len = ri->iri_frags[i].irf_len;
2441 	} while (++i < ri->iri_nfrags);
2442 
2443 	return (mh);
2444 }
2445 
2446 /*
2447  * Process one software descriptor
2448  */
2449 static struct mbuf *
2450 iflib_rxd_pkt_get(iflib_rxq_t rxq, if_rxd_info_t ri)
2451 {
2452 	struct if_rxsd sd;
2453 	struct mbuf *m;
2454 
2455 	/* should I merge this back in now that the two paths are basically duplicated? */
2456 	if (ri->iri_nfrags == 1 &&
2457 	    ri->iri_frags[0].irf_len <= MIN(IFLIB_RX_COPY_THRESH, MHLEN)) {
2458 		rxd_frag_to_sd(rxq, &ri->iri_frags[0], FALSE, &sd);
2459 		m = *sd.ifsd_m;
2460 		*sd.ifsd_m = NULL;
2461 		m_init(m, M_NOWAIT, MT_DATA, M_PKTHDR);
2462 #ifndef __NO_STRICT_ALIGNMENT
2463 		if (!IP_ALIGNED(m))
2464 			m->m_data += 2;
2465 #endif
2466 		memcpy(m->m_data, *sd.ifsd_cl, ri->iri_len);
2467 		m->m_len = ri->iri_frags[0].irf_len;
2468        } else {
2469 		m = assemble_segments(rxq, ri, &sd);
2470 	}
2471 	m->m_pkthdr.len = ri->iri_len;
2472 	m->m_pkthdr.rcvif = ri->iri_ifp;
2473 	m->m_flags |= ri->iri_flags;
2474 	m->m_pkthdr.ether_vtag = ri->iri_vtag;
2475 	m->m_pkthdr.flowid = ri->iri_flowid;
2476 	M_HASHTYPE_SET(m, ri->iri_rsstype);
2477 	m->m_pkthdr.csum_flags = ri->iri_csum_flags;
2478 	m->m_pkthdr.csum_data = ri->iri_csum_data;
2479 	return (m);
2480 }
2481 
2482 #if defined(INET6) || defined(INET)
2483 static void
2484 iflib_get_ip_forwarding(struct lro_ctrl *lc, bool *v4, bool *v6)
2485 {
2486 	CURVNET_SET(lc->ifp->if_vnet);
2487 #if defined(INET6)
2488 	*v6 = VNET(ip6_forwarding);
2489 #endif
2490 #if defined(INET)
2491 	*v4 = VNET(ipforwarding);
2492 #endif
2493 	CURVNET_RESTORE();
2494 }
2495 
2496 /*
2497  * Returns true if it's possible this packet could be LROed.
2498  * if it returns false, it is guaranteed that tcp_lro_rx()
2499  * would not return zero.
2500  */
2501 static bool
2502 iflib_check_lro_possible(struct mbuf *m, bool v4_forwarding, bool v6_forwarding)
2503 {
2504 	struct ether_header *eh;
2505 	uint16_t eh_type;
2506 
2507 	eh = mtod(m, struct ether_header *);
2508 	eh_type = ntohs(eh->ether_type);
2509 	switch (eh_type) {
2510 #if defined(INET6)
2511 		case ETHERTYPE_IPV6:
2512 			return !v6_forwarding;
2513 #endif
2514 #if defined (INET)
2515 		case ETHERTYPE_IP:
2516 			return !v4_forwarding;
2517 #endif
2518 	}
2519 
2520 	return false;
2521 }
2522 #else
2523 static void
2524 iflib_get_ip_forwarding(struct lro_ctrl *lc __unused, bool *v4 __unused, bool *v6 __unused)
2525 {
2526 }
2527 #endif
2528 
2529 static bool
2530 iflib_rxeof(iflib_rxq_t rxq, qidx_t budget)
2531 {
2532 	if_ctx_t ctx = rxq->ifr_ctx;
2533 	if_shared_ctx_t sctx = ctx->ifc_sctx;
2534 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2535 	int avail, i;
2536 	qidx_t *cidxp;
2537 	struct if_rxd_info ri;
2538 	int err, budget_left, rx_bytes, rx_pkts;
2539 	iflib_fl_t fl;
2540 	struct ifnet *ifp;
2541 	int lro_enabled;
2542 	bool lro_possible = false;
2543 	bool v4_forwarding, v6_forwarding;
2544 
2545 	/*
2546 	 * XXX early demux data packets so that if_input processing only handles
2547 	 * acks in interrupt context
2548 	 */
2549 	struct mbuf *m, *mh, *mt, *mf;
2550 
2551 	ifp = ctx->ifc_ifp;
2552 	mh = mt = NULL;
2553 	MPASS(budget > 0);
2554 	rx_pkts	= rx_bytes = 0;
2555 	if (sctx->isc_flags & IFLIB_HAS_RXCQ)
2556 		cidxp = &rxq->ifr_cq_cidx;
2557 	else
2558 		cidxp = &rxq->ifr_fl[0].ifl_cidx;
2559 	if ((avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget)) == 0) {
2560 		for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++)
2561 			__iflib_fl_refill_lt(ctx, fl, budget + 8);
2562 		DBG_COUNTER_INC(rx_unavail);
2563 		return (false);
2564 	}
2565 
2566 	for (budget_left = budget; (budget_left > 0) && (avail > 0); budget_left--, avail--) {
2567 		if (__predict_false(!CTX_ACTIVE(ctx))) {
2568 			DBG_COUNTER_INC(rx_ctx_inactive);
2569 			break;
2570 		}
2571 		/*
2572 		 * Reset client set fields to their default values
2573 		 */
2574 		rxd_info_zero(&ri);
2575 		ri.iri_qsidx = rxq->ifr_id;
2576 		ri.iri_cidx = *cidxp;
2577 		ri.iri_ifp = ifp;
2578 		ri.iri_frags = rxq->ifr_frags;
2579 		err = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri);
2580 
2581 		if (err)
2582 			goto err;
2583 		if (sctx->isc_flags & IFLIB_HAS_RXCQ) {
2584 			*cidxp = ri.iri_cidx;
2585 			/* Update our consumer index */
2586 			/* XXX NB: shurd - check if this is still safe */
2587 			while (rxq->ifr_cq_cidx >= scctx->isc_nrxd[0]) {
2588 				rxq->ifr_cq_cidx -= scctx->isc_nrxd[0];
2589 				rxq->ifr_cq_gen = 0;
2590 			}
2591 			/* was this only a completion queue message? */
2592 			if (__predict_false(ri.iri_nfrags == 0))
2593 				continue;
2594 		}
2595 		MPASS(ri.iri_nfrags != 0);
2596 		MPASS(ri.iri_len != 0);
2597 
2598 		/* will advance the cidx on the corresponding free lists */
2599 		m = iflib_rxd_pkt_get(rxq, &ri);
2600 		if (avail == 0 && budget_left)
2601 			avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget_left);
2602 
2603 		if (__predict_false(m == NULL)) {
2604 			DBG_COUNTER_INC(rx_mbuf_null);
2605 			continue;
2606 		}
2607 		/* imm_pkt: -- cxgb */
2608 		if (mh == NULL)
2609 			mh = mt = m;
2610 		else {
2611 			mt->m_nextpkt = m;
2612 			mt = m;
2613 		}
2614 	}
2615 	/* make sure that we can refill faster than drain */
2616 	for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++)
2617 		__iflib_fl_refill_lt(ctx, fl, budget + 8);
2618 
2619 	lro_enabled = (if_getcapenable(ifp) & IFCAP_LRO);
2620 	if (lro_enabled)
2621 		iflib_get_ip_forwarding(&rxq->ifr_lc, &v4_forwarding, &v6_forwarding);
2622 	mt = mf = NULL;
2623 	while (mh != NULL) {
2624 		m = mh;
2625 		mh = mh->m_nextpkt;
2626 		m->m_nextpkt = NULL;
2627 #ifndef __NO_STRICT_ALIGNMENT
2628 		if (!IP_ALIGNED(m) && (m = iflib_fixup_rx(m)) == NULL)
2629 			continue;
2630 #endif
2631 		rx_bytes += m->m_pkthdr.len;
2632 		rx_pkts++;
2633 #if defined(INET6) || defined(INET)
2634 		if (lro_enabled) {
2635 			if (!lro_possible) {
2636 				lro_possible = iflib_check_lro_possible(m, v4_forwarding, v6_forwarding);
2637 				if (lro_possible && mf != NULL) {
2638 					ifp->if_input(ifp, mf);
2639 					DBG_COUNTER_INC(rx_if_input);
2640 					mt = mf = NULL;
2641 				}
2642 			}
2643 			if ((m->m_pkthdr.csum_flags & (CSUM_L4_CALC|CSUM_L4_VALID)) ==
2644 			    (CSUM_L4_CALC|CSUM_L4_VALID)) {
2645 				if (lro_possible && tcp_lro_rx(&rxq->ifr_lc, m, 0) == 0)
2646 					continue;
2647 			}
2648 		}
2649 #endif
2650 		if (lro_possible) {
2651 			ifp->if_input(ifp, m);
2652 			DBG_COUNTER_INC(rx_if_input);
2653 			continue;
2654 		}
2655 
2656 		if (mf == NULL)
2657 			mf = m;
2658 		if (mt != NULL)
2659 			mt->m_nextpkt = m;
2660 		mt = m;
2661 	}
2662 	if (mf != NULL) {
2663 		ifp->if_input(ifp, mf);
2664 		DBG_COUNTER_INC(rx_if_input);
2665 	}
2666 
2667 	if_inc_counter(ifp, IFCOUNTER_IBYTES, rx_bytes);
2668 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, rx_pkts);
2669 
2670 	/*
2671 	 * Flush any outstanding LRO work
2672 	 */
2673 #if defined(INET6) || defined(INET)
2674 	tcp_lro_flush_all(&rxq->ifr_lc);
2675 #endif
2676 	if (avail)
2677 		return true;
2678 	return (iflib_rxd_avail(ctx, rxq, *cidxp, 1));
2679 err:
2680 	STATE_LOCK(ctx);
2681 	ctx->ifc_flags |= IFC_DO_RESET;
2682 	iflib_admin_intr_deferred(ctx);
2683 	STATE_UNLOCK(ctx);
2684 	return (false);
2685 }
2686 
2687 #define TXD_NOTIFY_COUNT(txq) (((txq)->ift_size / (txq)->ift_update_freq)-1)
2688 static inline qidx_t
2689 txq_max_db_deferred(iflib_txq_t txq, qidx_t in_use)
2690 {
2691 	qidx_t notify_count = TXD_NOTIFY_COUNT(txq);
2692 	qidx_t minthresh = txq->ift_size / 8;
2693 	if (in_use > 4*minthresh)
2694 		return (notify_count);
2695 	if (in_use > 2*minthresh)
2696 		return (notify_count >> 1);
2697 	if (in_use > minthresh)
2698 		return (notify_count >> 3);
2699 	return (0);
2700 }
2701 
2702 static inline qidx_t
2703 txq_max_rs_deferred(iflib_txq_t txq)
2704 {
2705 	qidx_t notify_count = TXD_NOTIFY_COUNT(txq);
2706 	qidx_t minthresh = txq->ift_size / 8;
2707 	if (txq->ift_in_use > 4*minthresh)
2708 		return (notify_count);
2709 	if (txq->ift_in_use > 2*minthresh)
2710 		return (notify_count >> 1);
2711 	if (txq->ift_in_use > minthresh)
2712 		return (notify_count >> 2);
2713 	return (2);
2714 }
2715 
2716 #define M_CSUM_FLAGS(m) ((m)->m_pkthdr.csum_flags)
2717 #define M_HAS_VLANTAG(m) (m->m_flags & M_VLANTAG)
2718 
2719 #define TXQ_MAX_DB_DEFERRED(txq, in_use) txq_max_db_deferred((txq), (in_use))
2720 #define TXQ_MAX_RS_DEFERRED(txq) txq_max_rs_deferred(txq)
2721 #define TXQ_MAX_DB_CONSUMED(size) (size >> 4)
2722 
2723 /* forward compatibility for cxgb */
2724 #define FIRST_QSET(ctx) 0
2725 #define NTXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_ntxqsets)
2726 #define NRXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_nrxqsets)
2727 #define QIDX(ctx, m) ((((m)->m_pkthdr.flowid & ctx->ifc_softc_ctx.isc_rss_table_mask) % NTXQSETS(ctx)) + FIRST_QSET(ctx))
2728 #define DESC_RECLAIMABLE(q) ((int)((q)->ift_processed - (q)->ift_cleaned - (q)->ift_ctx->ifc_softc_ctx.isc_tx_nsegments))
2729 
2730 /* XXX we should be setting this to something other than zero */
2731 #define RECLAIM_THRESH(ctx) ((ctx)->ifc_sctx->isc_tx_reclaim_thresh)
2732 #define MAX_TX_DESC(ctx) ((ctx)->ifc_softc_ctx.isc_tx_tso_segments_max)
2733 
2734 static inline bool
2735 iflib_txd_db_check(if_ctx_t ctx, iflib_txq_t txq, int ring, qidx_t in_use)
2736 {
2737 	qidx_t dbval, max;
2738 	bool rang;
2739 
2740 	rang = false;
2741 	max = TXQ_MAX_DB_DEFERRED(txq, in_use);
2742 	if (ring || txq->ift_db_pending >= max) {
2743 		dbval = txq->ift_npending ? txq->ift_npending : txq->ift_pidx;
2744 		ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, dbval);
2745 		txq->ift_db_pending = txq->ift_npending = 0;
2746 		rang = true;
2747 	}
2748 	return (rang);
2749 }
2750 
2751 #ifdef PKT_DEBUG
2752 static void
2753 print_pkt(if_pkt_info_t pi)
2754 {
2755 	printf("pi len:  %d qsidx: %d nsegs: %d ndescs: %d flags: %x pidx: %d\n",
2756 	       pi->ipi_len, pi->ipi_qsidx, pi->ipi_nsegs, pi->ipi_ndescs, pi->ipi_flags, pi->ipi_pidx);
2757 	printf("pi new_pidx: %d csum_flags: %lx tso_segsz: %d mflags: %x vtag: %d\n",
2758 	       pi->ipi_new_pidx, pi->ipi_csum_flags, pi->ipi_tso_segsz, pi->ipi_mflags, pi->ipi_vtag);
2759 	printf("pi etype: %d ehdrlen: %d ip_hlen: %d ipproto: %d\n",
2760 	       pi->ipi_etype, pi->ipi_ehdrlen, pi->ipi_ip_hlen, pi->ipi_ipproto);
2761 }
2762 #endif
2763 
2764 #define IS_TSO4(pi) ((pi)->ipi_csum_flags & CSUM_IP_TSO)
2765 #define IS_TSO6(pi) ((pi)->ipi_csum_flags & CSUM_IP6_TSO)
2766 
2767 static int
2768 iflib_parse_header(iflib_txq_t txq, if_pkt_info_t pi, struct mbuf **mp)
2769 {
2770 	if_shared_ctx_t sctx = txq->ift_ctx->ifc_sctx;
2771 	struct ether_vlan_header *eh;
2772 	struct mbuf *m, *n;
2773 
2774 	n = m = *mp;
2775 	if ((sctx->isc_flags & IFLIB_NEED_SCRATCH) &&
2776 	    M_WRITABLE(m) == 0) {
2777 		if ((m = m_dup(m, M_NOWAIT)) == NULL) {
2778 			return (ENOMEM);
2779 		} else {
2780 			m_freem(*mp);
2781 			n = *mp = m;
2782 		}
2783 	}
2784 
2785 	/*
2786 	 * Determine where frame payload starts.
2787 	 * Jump over vlan headers if already present,
2788 	 * helpful for QinQ too.
2789 	 */
2790 	if (__predict_false(m->m_len < sizeof(*eh))) {
2791 		txq->ift_pullups++;
2792 		if (__predict_false((m = m_pullup(m, sizeof(*eh))) == NULL))
2793 			return (ENOMEM);
2794 	}
2795 	eh = mtod(m, struct ether_vlan_header *);
2796 	if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2797 		pi->ipi_etype = ntohs(eh->evl_proto);
2798 		pi->ipi_ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
2799 	} else {
2800 		pi->ipi_etype = ntohs(eh->evl_encap_proto);
2801 		pi->ipi_ehdrlen = ETHER_HDR_LEN;
2802 	}
2803 
2804 	switch (pi->ipi_etype) {
2805 #ifdef INET
2806 	case ETHERTYPE_IP:
2807 	{
2808 		struct ip *ip = NULL;
2809 		struct tcphdr *th = NULL;
2810 		int minthlen;
2811 
2812 		minthlen = min(m->m_pkthdr.len, pi->ipi_ehdrlen + sizeof(*ip) + sizeof(*th));
2813 		if (__predict_false(m->m_len < minthlen)) {
2814 			/*
2815 			 * if this code bloat is causing too much of a hit
2816 			 * move it to a separate function and mark it noinline
2817 			 */
2818 			if (m->m_len == pi->ipi_ehdrlen) {
2819 				n = m->m_next;
2820 				MPASS(n);
2821 				if (n->m_len >= sizeof(*ip))  {
2822 					ip = (struct ip *)n->m_data;
2823 					if (n->m_len >= (ip->ip_hl << 2) + sizeof(*th))
2824 						th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2825 				} else {
2826 					txq->ift_pullups++;
2827 					if (__predict_false((m = m_pullup(m, minthlen)) == NULL))
2828 						return (ENOMEM);
2829 					ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
2830 				}
2831 			} else {
2832 				txq->ift_pullups++;
2833 				if (__predict_false((m = m_pullup(m, minthlen)) == NULL))
2834 					return (ENOMEM);
2835 				ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
2836 				if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th))
2837 					th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2838 			}
2839 		} else {
2840 			ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
2841 			if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th))
2842 				th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2843 		}
2844 		pi->ipi_ip_hlen = ip->ip_hl << 2;
2845 		pi->ipi_ipproto = ip->ip_p;
2846 		pi->ipi_flags |= IPI_TX_IPV4;
2847 
2848 		if ((sctx->isc_flags & IFLIB_NEED_ZERO_CSUM) && (pi->ipi_csum_flags & CSUM_IP))
2849                        ip->ip_sum = 0;
2850 
2851 		if (IS_TSO4(pi)) {
2852 			if (pi->ipi_ipproto == IPPROTO_TCP) {
2853 				if (__predict_false(th == NULL)) {
2854 					txq->ift_pullups++;
2855 					if (__predict_false((m = m_pullup(m, (ip->ip_hl << 2) + sizeof(*th))) == NULL))
2856 						return (ENOMEM);
2857 					th = (struct tcphdr *)((caddr_t)ip + pi->ipi_ip_hlen);
2858 				}
2859 				pi->ipi_tcp_hflags = th->th_flags;
2860 				pi->ipi_tcp_hlen = th->th_off << 2;
2861 				pi->ipi_tcp_seq = th->th_seq;
2862 			}
2863 			if (__predict_false(ip->ip_p != IPPROTO_TCP))
2864 				return (ENXIO);
2865 			th->th_sum = in_pseudo(ip->ip_src.s_addr,
2866 					       ip->ip_dst.s_addr, htons(IPPROTO_TCP));
2867 			pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz;
2868 			if (sctx->isc_flags & IFLIB_TSO_INIT_IP) {
2869 				ip->ip_sum = 0;
2870 				ip->ip_len = htons(pi->ipi_ip_hlen + pi->ipi_tcp_hlen + pi->ipi_tso_segsz);
2871 			}
2872 		}
2873 		break;
2874 	}
2875 #endif
2876 #ifdef INET6
2877 	case ETHERTYPE_IPV6:
2878 	{
2879 		struct ip6_hdr *ip6 = (struct ip6_hdr *)(m->m_data + pi->ipi_ehdrlen);
2880 		struct tcphdr *th;
2881 		pi->ipi_ip_hlen = sizeof(struct ip6_hdr);
2882 
2883 		if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) {
2884 			if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) == NULL))
2885 				return (ENOMEM);
2886 		}
2887 		th = (struct tcphdr *)((caddr_t)ip6 + pi->ipi_ip_hlen);
2888 
2889 		/* XXX-BZ this will go badly in case of ext hdrs. */
2890 		pi->ipi_ipproto = ip6->ip6_nxt;
2891 		pi->ipi_flags |= IPI_TX_IPV6;
2892 
2893 		if (IS_TSO6(pi)) {
2894 			if (pi->ipi_ipproto == IPPROTO_TCP) {
2895 				if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) {
2896 					if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) == NULL))
2897 						return (ENOMEM);
2898 				}
2899 				pi->ipi_tcp_hflags = th->th_flags;
2900 				pi->ipi_tcp_hlen = th->th_off << 2;
2901 			}
2902 
2903 			if (__predict_false(ip6->ip6_nxt != IPPROTO_TCP))
2904 				return (ENXIO);
2905 			/*
2906 			 * The corresponding flag is set by the stack in the IPv4
2907 			 * TSO case, but not in IPv6 (at least in FreeBSD 10.2).
2908 			 * So, set it here because the rest of the flow requires it.
2909 			 */
2910 			pi->ipi_csum_flags |= CSUM_TCP_IPV6;
2911 			th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
2912 			pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz;
2913 		}
2914 		break;
2915 	}
2916 #endif
2917 	default:
2918 		pi->ipi_csum_flags &= ~CSUM_OFFLOAD;
2919 		pi->ipi_ip_hlen = 0;
2920 		break;
2921 	}
2922 	*mp = m;
2923 
2924 	return (0);
2925 }
2926 
2927 static  __noinline  struct mbuf *
2928 collapse_pkthdr(struct mbuf *m0)
2929 {
2930 	struct mbuf *m, *m_next, *tmp;
2931 
2932 	m = m0;
2933 	m_next = m->m_next;
2934 	while (m_next != NULL && m_next->m_len == 0) {
2935 		m = m_next;
2936 		m->m_next = NULL;
2937 		m_free(m);
2938 		m_next = m_next->m_next;
2939 	}
2940 	m = m0;
2941 	m->m_next = m_next;
2942 	if ((m_next->m_flags & M_EXT) == 0) {
2943 		m = m_defrag(m, M_NOWAIT);
2944 	} else {
2945 		tmp = m_next->m_next;
2946 		memcpy(m_next, m, MPKTHSIZE);
2947 		m = m_next;
2948 		m->m_next = tmp;
2949 	}
2950 	return (m);
2951 }
2952 
2953 /*
2954  * If dodgy hardware rejects the scatter gather chain we've handed it
2955  * we'll need to remove the mbuf chain from ifsg_m[] before we can add the
2956  * m_defrag'd mbufs
2957  */
2958 static __noinline struct mbuf *
2959 iflib_remove_mbuf(iflib_txq_t txq)
2960 {
2961 	int ntxd, i, pidx;
2962 	struct mbuf *m, *mh, **ifsd_m;
2963 
2964 	pidx = txq->ift_pidx;
2965 	ifsd_m = txq->ift_sds.ifsd_m;
2966 	ntxd = txq->ift_size;
2967 	mh = m = ifsd_m[pidx];
2968 	ifsd_m[pidx] = NULL;
2969 #if MEMORY_LOGGING
2970 	txq->ift_dequeued++;
2971 #endif
2972 	i = 1;
2973 
2974 	while (m) {
2975 		ifsd_m[(pidx + i) & (ntxd -1)] = NULL;
2976 #if MEMORY_LOGGING
2977 		txq->ift_dequeued++;
2978 #endif
2979 		m = m->m_next;
2980 		i++;
2981 	}
2982 	return (mh);
2983 }
2984 
2985 static int
2986 iflib_busdma_load_mbuf_sg(iflib_txq_t txq, bus_dma_tag_t tag, bus_dmamap_t map,
2987 			  struct mbuf **m0, bus_dma_segment_t *segs, int *nsegs,
2988 			  int max_segs, int flags)
2989 {
2990 	if_ctx_t ctx;
2991 	if_shared_ctx_t		sctx;
2992 	if_softc_ctx_t		scctx;
2993 	int i, next, pidx, err, ntxd, count;
2994 	struct mbuf *m, *tmp, **ifsd_m;
2995 
2996 	m = *m0;
2997 
2998 	/*
2999 	 * Please don't ever do this
3000 	 */
3001 	if (__predict_false(m->m_len == 0))
3002 		*m0 = m = collapse_pkthdr(m);
3003 
3004 	ctx = txq->ift_ctx;
3005 	sctx = ctx->ifc_sctx;
3006 	scctx = &ctx->ifc_softc_ctx;
3007 	ifsd_m = txq->ift_sds.ifsd_m;
3008 	ntxd = txq->ift_size;
3009 	pidx = txq->ift_pidx;
3010 	if (map != NULL) {
3011 		uint8_t *ifsd_flags = txq->ift_sds.ifsd_flags;
3012 
3013 		err = bus_dmamap_load_mbuf_sg(tag, map,
3014 					      *m0, segs, nsegs, BUS_DMA_NOWAIT);
3015 		if (err)
3016 			return (err);
3017 		ifsd_flags[pidx] |= TX_SW_DESC_MAPPED;
3018 		count = 0;
3019 		m = *m0;
3020 		do {
3021 			if (__predict_false(m->m_len <= 0)) {
3022 				tmp = m;
3023 				m = m->m_next;
3024 				tmp->m_next = NULL;
3025 				m_free(tmp);
3026 				continue;
3027 			}
3028 			m = m->m_next;
3029 			count++;
3030 		} while (m != NULL);
3031 		if (count > *nsegs) {
3032 			ifsd_m[pidx] = *m0;
3033 			ifsd_m[pidx]->m_flags |= M_TOOBIG;
3034 			return (0);
3035 		}
3036 		m = *m0;
3037 		count = 0;
3038 		do {
3039 			next = (pidx + count) & (ntxd-1);
3040 			MPASS(ifsd_m[next] == NULL);
3041 			ifsd_m[next] = m;
3042 			count++;
3043 			tmp = m;
3044 			m = m->m_next;
3045 		} while (m != NULL);
3046 	} else {
3047 		int buflen, sgsize, maxsegsz, max_sgsize;
3048 		vm_offset_t vaddr;
3049 		vm_paddr_t curaddr;
3050 
3051 		count = i = 0;
3052 		m = *m0;
3053 		if (m->m_pkthdr.csum_flags & CSUM_TSO)
3054 			maxsegsz = scctx->isc_tx_tso_segsize_max;
3055 		else
3056 			maxsegsz = sctx->isc_tx_maxsegsize;
3057 
3058 		do {
3059 			if (__predict_false(m->m_len <= 0)) {
3060 				tmp = m;
3061 				m = m->m_next;
3062 				tmp->m_next = NULL;
3063 				m_free(tmp);
3064 				continue;
3065 			}
3066 			buflen = m->m_len;
3067 			vaddr = (vm_offset_t)m->m_data;
3068 			/*
3069 			 * see if we can't be smarter about physically
3070 			 * contiguous mappings
3071 			 */
3072 			next = (pidx + count) & (ntxd-1);
3073 			MPASS(ifsd_m[next] == NULL);
3074 #if MEMORY_LOGGING
3075 			txq->ift_enqueued++;
3076 #endif
3077 			ifsd_m[next] = m;
3078 			while (buflen > 0) {
3079 				if (i >= max_segs)
3080 					goto err;
3081 				max_sgsize = MIN(buflen, maxsegsz);
3082 				curaddr = pmap_kextract(vaddr);
3083 				sgsize = PAGE_SIZE - (curaddr & PAGE_MASK);
3084 				sgsize = MIN(sgsize, max_sgsize);
3085 				segs[i].ds_addr = curaddr;
3086 				segs[i].ds_len = sgsize;
3087 				vaddr += sgsize;
3088 				buflen -= sgsize;
3089 				i++;
3090 			}
3091 			count++;
3092 			tmp = m;
3093 			m = m->m_next;
3094 		} while (m != NULL);
3095 		*nsegs = i;
3096 	}
3097 	return (0);
3098 err:
3099 	*m0 = iflib_remove_mbuf(txq);
3100 	return (EFBIG);
3101 }
3102 
3103 static inline caddr_t
3104 calc_next_txd(iflib_txq_t txq, int cidx, uint8_t qid)
3105 {
3106 	qidx_t size;
3107 	int ntxd;
3108 	caddr_t start, end, cur, next;
3109 
3110 	ntxd = txq->ift_size;
3111 	size = txq->ift_txd_size[qid];
3112 	start = txq->ift_ifdi[qid].idi_vaddr;
3113 
3114 	if (__predict_false(size == 0))
3115 		return (start);
3116 	cur = start + size*cidx;
3117 	end = start + size*ntxd;
3118 	next = CACHE_PTR_NEXT(cur);
3119 	return (next < end ? next : start);
3120 }
3121 
3122 /*
3123  * Pad an mbuf to ensure a minimum ethernet frame size.
3124  * min_frame_size is the frame size (less CRC) to pad the mbuf to
3125  */
3126 static __noinline int
3127 iflib_ether_pad(device_t dev, struct mbuf **m_head, uint16_t min_frame_size)
3128 {
3129 	/*
3130 	 * 18 is enough bytes to pad an ARP packet to 46 bytes, and
3131 	 * and ARP message is the smallest common payload I can think of
3132 	 */
3133 	static char pad[18];	/* just zeros */
3134 	int n;
3135 	struct mbuf *new_head;
3136 
3137 	if (!M_WRITABLE(*m_head)) {
3138 		new_head = m_dup(*m_head, M_NOWAIT);
3139 		if (new_head == NULL) {
3140 			m_freem(*m_head);
3141 			device_printf(dev, "cannot pad short frame, m_dup() failed");
3142 			DBG_COUNTER_INC(encap_pad_mbuf_fail);
3143 			return ENOMEM;
3144 		}
3145 		m_freem(*m_head);
3146 		*m_head = new_head;
3147 	}
3148 
3149 	for (n = min_frame_size - (*m_head)->m_pkthdr.len;
3150 	     n > 0; n -= sizeof(pad))
3151 		if (!m_append(*m_head, min(n, sizeof(pad)), pad))
3152 			break;
3153 
3154 	if (n > 0) {
3155 		m_freem(*m_head);
3156 		device_printf(dev, "cannot pad short frame\n");
3157 		DBG_COUNTER_INC(encap_pad_mbuf_fail);
3158 		return (ENOBUFS);
3159 	}
3160 
3161 	return 0;
3162 }
3163 
3164 static int
3165 iflib_encap(iflib_txq_t txq, struct mbuf **m_headp)
3166 {
3167 	if_ctx_t		ctx;
3168 	if_shared_ctx_t		sctx;
3169 	if_softc_ctx_t		scctx;
3170 	bus_dma_segment_t	*segs;
3171 	struct mbuf		*m_head;
3172 	void			*next_txd;
3173 	bus_dmamap_t		map;
3174 	struct if_pkt_info	pi;
3175 	int remap = 0;
3176 	int err, nsegs, ndesc, max_segs, pidx, cidx, next, ntxd;
3177 	bus_dma_tag_t desc_tag;
3178 
3179 	segs = txq->ift_segs;
3180 	ctx = txq->ift_ctx;
3181 	sctx = ctx->ifc_sctx;
3182 	scctx = &ctx->ifc_softc_ctx;
3183 	segs = txq->ift_segs;
3184 	ntxd = txq->ift_size;
3185 	m_head = *m_headp;
3186 	map = NULL;
3187 
3188 	/*
3189 	 * If we're doing TSO the next descriptor to clean may be quite far ahead
3190 	 */
3191 	cidx = txq->ift_cidx;
3192 	pidx = txq->ift_pidx;
3193 	if (ctx->ifc_flags & IFC_PREFETCH) {
3194 		next = (cidx + CACHE_PTR_INCREMENT) & (ntxd-1);
3195 		if (!(ctx->ifc_flags & IFLIB_HAS_TXCQ)) {
3196 			next_txd = calc_next_txd(txq, cidx, 0);
3197 			prefetch(next_txd);
3198 		}
3199 
3200 		/* prefetch the next cache line of mbuf pointers and flags */
3201 		prefetch(&txq->ift_sds.ifsd_m[next]);
3202 		if (txq->ift_sds.ifsd_map != NULL) {
3203 			prefetch(&txq->ift_sds.ifsd_map[next]);
3204 			next = (cidx + CACHE_LINE_SIZE) & (ntxd-1);
3205 			prefetch(&txq->ift_sds.ifsd_flags[next]);
3206 		}
3207 	} else if (txq->ift_sds.ifsd_map != NULL)
3208 		map = txq->ift_sds.ifsd_map[pidx];
3209 
3210 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3211 		desc_tag = txq->ift_tso_desc_tag;
3212 		max_segs = scctx->isc_tx_tso_segments_max;
3213 	} else {
3214 		desc_tag = txq->ift_desc_tag;
3215 		max_segs = scctx->isc_tx_nsegments;
3216 	}
3217 	if ((sctx->isc_flags & IFLIB_NEED_ETHER_PAD) &&
3218 	    __predict_false(m_head->m_pkthdr.len < scctx->isc_min_frame_size)) {
3219 		err = iflib_ether_pad(ctx->ifc_dev, m_headp, scctx->isc_min_frame_size);
3220 		if (err)
3221 			return err;
3222 	}
3223 	m_head = *m_headp;
3224 
3225 	pkt_info_zero(&pi);
3226 	pi.ipi_mflags = (m_head->m_flags & (M_VLANTAG|M_BCAST|M_MCAST));
3227 	pi.ipi_pidx = pidx;
3228 	pi.ipi_qsidx = txq->ift_id;
3229 	pi.ipi_len = m_head->m_pkthdr.len;
3230 	pi.ipi_csum_flags = m_head->m_pkthdr.csum_flags;
3231 	pi.ipi_vtag = (m_head->m_flags & M_VLANTAG) ? m_head->m_pkthdr.ether_vtag : 0;
3232 
3233 	/* deliberate bitwise OR to make one condition */
3234 	if (__predict_true((pi.ipi_csum_flags | pi.ipi_vtag))) {
3235 		if (__predict_false((err = iflib_parse_header(txq, &pi, m_headp)) != 0))
3236 			return (err);
3237 		m_head = *m_headp;
3238 	}
3239 
3240 retry:
3241 	err = iflib_busdma_load_mbuf_sg(txq, desc_tag, map, m_headp, segs, &nsegs, max_segs, BUS_DMA_NOWAIT);
3242 defrag:
3243 	if (__predict_false(err)) {
3244 		switch (err) {
3245 		case EFBIG:
3246 			/* try collapse once and defrag once */
3247 			if (remap == 0) {
3248 				m_head = m_collapse(*m_headp, M_NOWAIT, max_segs);
3249 				/* try defrag if collapsing fails */
3250 				if (m_head == NULL)
3251 					remap++;
3252 			}
3253 			if (remap == 1)
3254 				m_head = m_defrag(*m_headp, M_NOWAIT);
3255 			remap++;
3256 			if (__predict_false(m_head == NULL))
3257 				goto defrag_failed;
3258 			txq->ift_mbuf_defrag++;
3259 			*m_headp = m_head;
3260 			goto retry;
3261 			break;
3262 		case ENOMEM:
3263 			txq->ift_no_tx_dma_setup++;
3264 			break;
3265 		default:
3266 			txq->ift_no_tx_dma_setup++;
3267 			m_freem(*m_headp);
3268 			DBG_COUNTER_INC(tx_frees);
3269 			*m_headp = NULL;
3270 			break;
3271 		}
3272 		txq->ift_map_failed++;
3273 		DBG_COUNTER_INC(encap_load_mbuf_fail);
3274 		return (err);
3275 	}
3276 
3277 	/*
3278 	 * XXX assumes a 1 to 1 relationship between segments and
3279 	 *        descriptors - this does not hold true on all drivers, e.g.
3280 	 *        cxgb
3281 	 */
3282 	if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) {
3283 		txq->ift_no_desc_avail++;
3284 		if (map != NULL)
3285 			bus_dmamap_unload(desc_tag, map);
3286 		DBG_COUNTER_INC(encap_txq_avail_fail);
3287 		if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0)
3288 			GROUPTASK_ENQUEUE(&txq->ift_task);
3289 		return (ENOBUFS);
3290 	}
3291 	/*
3292 	 * On Intel cards we can greatly reduce the number of TX interrupts
3293 	 * we see by only setting report status on every Nth descriptor.
3294 	 * However, this also means that the driver will need to keep track
3295 	 * of the descriptors that RS was set on to check them for the DD bit.
3296 	 */
3297 	txq->ift_rs_pending += nsegs + 1;
3298 	if (txq->ift_rs_pending > TXQ_MAX_RS_DEFERRED(txq) ||
3299 	     iflib_no_tx_batch || (TXQ_AVAIL(txq) - nsegs - 1) <= MAX_TX_DESC(ctx)) {
3300 		pi.ipi_flags |= IPI_TX_INTR;
3301 		txq->ift_rs_pending = 0;
3302 	}
3303 
3304 	pi.ipi_segs = segs;
3305 	pi.ipi_nsegs = nsegs;
3306 
3307 	MPASS(pidx >= 0 && pidx < txq->ift_size);
3308 #ifdef PKT_DEBUG
3309 	print_pkt(&pi);
3310 #endif
3311 	if (map != NULL)
3312 		bus_dmamap_sync(desc_tag, map, BUS_DMASYNC_PREWRITE);
3313 	if ((err = ctx->isc_txd_encap(ctx->ifc_softc, &pi)) == 0) {
3314 		if (map != NULL)
3315 			bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map,
3316 					BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
3317 		DBG_COUNTER_INC(tx_encap);
3318 		MPASS(pi.ipi_new_pidx < txq->ift_size);
3319 
3320 		ndesc = pi.ipi_new_pidx - pi.ipi_pidx;
3321 		if (pi.ipi_new_pidx < pi.ipi_pidx) {
3322 			ndesc += txq->ift_size;
3323 			txq->ift_gen = 1;
3324 		}
3325 		/*
3326 		 * drivers can need as many as
3327 		 * two sentinels
3328 		 */
3329 		MPASS(ndesc <= pi.ipi_nsegs + 2);
3330 		MPASS(pi.ipi_new_pidx != pidx);
3331 		MPASS(ndesc > 0);
3332 		txq->ift_in_use += ndesc;
3333 
3334 		/*
3335 		 * We update the last software descriptor again here because there may
3336 		 * be a sentinel and/or there may be more mbufs than segments
3337 		 */
3338 		txq->ift_pidx = pi.ipi_new_pidx;
3339 		txq->ift_npending += pi.ipi_ndescs;
3340 	} else {
3341 		*m_headp = m_head = iflib_remove_mbuf(txq);
3342 		if (err == EFBIG) {
3343 			txq->ift_txd_encap_efbig++;
3344 			if (remap < 2) {
3345 				remap = 1;
3346 				goto defrag;
3347 			}
3348 		}
3349 		DBG_COUNTER_INC(encap_txd_encap_fail);
3350 		goto defrag_failed;
3351 	}
3352 	return (err);
3353 
3354 defrag_failed:
3355 	txq->ift_mbuf_defrag_failed++;
3356 	txq->ift_map_failed++;
3357 	m_freem(*m_headp);
3358 	DBG_COUNTER_INC(tx_frees);
3359 	*m_headp = NULL;
3360 	return (ENOMEM);
3361 }
3362 
3363 static void
3364 iflib_tx_desc_free(iflib_txq_t txq, int n)
3365 {
3366 	int hasmap;
3367 	uint32_t qsize, cidx, mask, gen;
3368 	struct mbuf *m, **ifsd_m;
3369 	uint8_t *ifsd_flags;
3370 	bus_dmamap_t *ifsd_map;
3371 	bool do_prefetch;
3372 
3373 	cidx = txq->ift_cidx;
3374 	gen = txq->ift_gen;
3375 	qsize = txq->ift_size;
3376 	mask = qsize-1;
3377 	hasmap = txq->ift_sds.ifsd_map != NULL;
3378 	ifsd_flags = txq->ift_sds.ifsd_flags;
3379 	ifsd_m = txq->ift_sds.ifsd_m;
3380 	ifsd_map = txq->ift_sds.ifsd_map;
3381 	do_prefetch = (txq->ift_ctx->ifc_flags & IFC_PREFETCH);
3382 
3383 	while (n--) {
3384 		if (do_prefetch) {
3385 			prefetch(ifsd_m[(cidx + 3) & mask]);
3386 			prefetch(ifsd_m[(cidx + 4) & mask]);
3387 		}
3388 		if (ifsd_m[cidx] != NULL) {
3389 			prefetch(&ifsd_m[(cidx + CACHE_PTR_INCREMENT) & mask]);
3390 			prefetch(&ifsd_flags[(cidx + CACHE_PTR_INCREMENT) & mask]);
3391 			if (hasmap && (ifsd_flags[cidx] & TX_SW_DESC_MAPPED)) {
3392 				/*
3393 				 * does it matter if it's not the TSO tag? If so we'll
3394 				 * have to add the type to flags
3395 				 */
3396 				bus_dmamap_unload(txq->ift_desc_tag, ifsd_map[cidx]);
3397 				ifsd_flags[cidx] &= ~TX_SW_DESC_MAPPED;
3398 			}
3399 			if ((m = ifsd_m[cidx]) != NULL) {
3400 				/* XXX we don't support any drivers that batch packets yet */
3401 				MPASS(m->m_nextpkt == NULL);
3402 				/* if the number of clusters exceeds the number of segments
3403 				 * there won't be space on the ring to save a pointer to each
3404 				 * cluster so we simply free the list here
3405 				 */
3406 				if (m->m_flags & M_TOOBIG) {
3407 					m_freem(m);
3408 				} else {
3409 					m_free(m);
3410 				}
3411 				ifsd_m[cidx] = NULL;
3412 #if MEMORY_LOGGING
3413 				txq->ift_dequeued++;
3414 #endif
3415 				DBG_COUNTER_INC(tx_frees);
3416 			}
3417 		}
3418 		if (__predict_false(++cidx == qsize)) {
3419 			cidx = 0;
3420 			gen = 0;
3421 		}
3422 	}
3423 	txq->ift_cidx = cidx;
3424 	txq->ift_gen = gen;
3425 }
3426 
3427 static __inline int
3428 iflib_completed_tx_reclaim(iflib_txq_t txq, int thresh)
3429 {
3430 	int reclaim;
3431 	if_ctx_t ctx = txq->ift_ctx;
3432 
3433 	KASSERT(thresh >= 0, ("invalid threshold to reclaim"));
3434 	MPASS(thresh /*+ MAX_TX_DESC(txq->ift_ctx) */ < txq->ift_size);
3435 
3436 	/*
3437 	 * Need a rate-limiting check so that this isn't called every time
3438 	 */
3439 	iflib_tx_credits_update(ctx, txq);
3440 	reclaim = DESC_RECLAIMABLE(txq);
3441 
3442 	if (reclaim <= thresh /* + MAX_TX_DESC(txq->ift_ctx) */) {
3443 #ifdef INVARIANTS
3444 		if (iflib_verbose_debug) {
3445 			printf("%s processed=%ju cleaned=%ju tx_nsegments=%d reclaim=%d thresh=%d\n", __FUNCTION__,
3446 			       txq->ift_processed, txq->ift_cleaned, txq->ift_ctx->ifc_softc_ctx.isc_tx_nsegments,
3447 			       reclaim, thresh);
3448 
3449 		}
3450 #endif
3451 		return (0);
3452 	}
3453 	iflib_tx_desc_free(txq, reclaim);
3454 	txq->ift_cleaned += reclaim;
3455 	txq->ift_in_use -= reclaim;
3456 
3457 	return (reclaim);
3458 }
3459 
3460 static struct mbuf **
3461 _ring_peek_one(struct ifmp_ring *r, int cidx, int offset, int remaining)
3462 {
3463 	int next, size;
3464 	struct mbuf **items;
3465 
3466 	size = r->size;
3467 	next = (cidx + CACHE_PTR_INCREMENT) & (size-1);
3468 	items = __DEVOLATILE(struct mbuf **, &r->items[0]);
3469 
3470 	prefetch(items[(cidx + offset) & (size-1)]);
3471 	if (remaining > 1) {
3472 		prefetch2cachelines(&items[next]);
3473 		prefetch2cachelines(items[(cidx + offset + 1) & (size-1)]);
3474 		prefetch2cachelines(items[(cidx + offset + 2) & (size-1)]);
3475 		prefetch2cachelines(items[(cidx + offset + 3) & (size-1)]);
3476 	}
3477 	return (__DEVOLATILE(struct mbuf **, &r->items[(cidx + offset) & (size-1)]));
3478 }
3479 
3480 static void
3481 iflib_txq_check_drain(iflib_txq_t txq, int budget)
3482 {
3483 
3484 	ifmp_ring_check_drainage(txq->ift_br, budget);
3485 }
3486 
3487 static uint32_t
3488 iflib_txq_can_drain(struct ifmp_ring *r)
3489 {
3490 	iflib_txq_t txq = r->cookie;
3491 	if_ctx_t ctx = txq->ift_ctx;
3492 
3493 	return ((TXQ_AVAIL(txq) > MAX_TX_DESC(ctx) + 2) ||
3494 		ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false));
3495 }
3496 
3497 static uint32_t
3498 iflib_txq_drain(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx)
3499 {
3500 	iflib_txq_t txq = r->cookie;
3501 	if_ctx_t ctx = txq->ift_ctx;
3502 	struct ifnet *ifp = ctx->ifc_ifp;
3503 	struct mbuf **mp, *m;
3504 	int i, count, consumed, pkt_sent, bytes_sent, mcast_sent, avail;
3505 	int reclaimed, err, in_use_prev, desc_used;
3506 	bool do_prefetch, ring, rang;
3507 
3508 	if (__predict_false(!(if_getdrvflags(ifp) & IFF_DRV_RUNNING) ||
3509 			    !LINK_ACTIVE(ctx))) {
3510 		DBG_COUNTER_INC(txq_drain_notready);
3511 		return (0);
3512 	}
3513 	reclaimed = iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx));
3514 	rang = iflib_txd_db_check(ctx, txq, reclaimed, txq->ift_in_use);
3515 	avail = IDXDIFF(pidx, cidx, r->size);
3516 	if (__predict_false(ctx->ifc_flags & IFC_QFLUSH)) {
3517 		DBG_COUNTER_INC(txq_drain_flushing);
3518 		for (i = 0; i < avail; i++) {
3519 			m_free(r->items[(cidx + i) & (r->size-1)]);
3520 			r->items[(cidx + i) & (r->size-1)] = NULL;
3521 		}
3522 		return (avail);
3523 	}
3524 
3525 	if (__predict_false(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) {
3526 		txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3527 		CALLOUT_LOCK(txq);
3528 		callout_stop(&txq->ift_timer);
3529 		CALLOUT_UNLOCK(txq);
3530 		DBG_COUNTER_INC(txq_drain_oactive);
3531 		return (0);
3532 	}
3533 	if (reclaimed)
3534 		txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3535 	consumed = mcast_sent = bytes_sent = pkt_sent = 0;
3536 	count = MIN(avail, TX_BATCH_SIZE);
3537 #ifdef INVARIANTS
3538 	if (iflib_verbose_debug)
3539 		printf("%s avail=%d ifc_flags=%x txq_avail=%d ", __FUNCTION__,
3540 		       avail, ctx->ifc_flags, TXQ_AVAIL(txq));
3541 #endif
3542 	do_prefetch = (ctx->ifc_flags & IFC_PREFETCH);
3543 	avail = TXQ_AVAIL(txq);
3544 	for (desc_used = i = 0; i < count && avail > MAX_TX_DESC(ctx) + 2; i++) {
3545 		int pidx_prev, rem = do_prefetch ? count - i : 0;
3546 
3547 		mp = _ring_peek_one(r, cidx, i, rem);
3548 		MPASS(mp != NULL && *mp != NULL);
3549 		if (__predict_false(*mp == (struct mbuf *)txq)) {
3550 			consumed++;
3551 			reclaimed++;
3552 			continue;
3553 		}
3554 		in_use_prev = txq->ift_in_use;
3555 		pidx_prev = txq->ift_pidx;
3556 		err = iflib_encap(txq, mp);
3557 		if (__predict_false(err)) {
3558 			DBG_COUNTER_INC(txq_drain_encapfail);
3559 			/* no room - bail out */
3560 			if (err == ENOBUFS)
3561 				break;
3562 			consumed++;
3563 			DBG_COUNTER_INC(txq_drain_encapfail);
3564 			/* we can't send this packet - skip it */
3565 			continue;
3566 		}
3567 		consumed++;
3568 		pkt_sent++;
3569 		m = *mp;
3570 		DBG_COUNTER_INC(tx_sent);
3571 		bytes_sent += m->m_pkthdr.len;
3572 		mcast_sent += !!(m->m_flags & M_MCAST);
3573 		avail = TXQ_AVAIL(txq);
3574 
3575 		txq->ift_db_pending += (txq->ift_in_use - in_use_prev);
3576 		desc_used += (txq->ift_in_use - in_use_prev);
3577 		ETHER_BPF_MTAP(ifp, m);
3578 		if (__predict_false(!(ifp->if_drv_flags & IFF_DRV_RUNNING)))
3579 			break;
3580 		rang = iflib_txd_db_check(ctx, txq, false, in_use_prev);
3581 	}
3582 
3583 	/* deliberate use of bitwise or to avoid gratuitous short-circuit */
3584 	ring = rang ? false  : (iflib_min_tx_latency | err) || (TXQ_AVAIL(txq) < MAX_TX_DESC(ctx));
3585 	iflib_txd_db_check(ctx, txq, ring, txq->ift_in_use);
3586 	if_inc_counter(ifp, IFCOUNTER_OBYTES, bytes_sent);
3587 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, pkt_sent);
3588 	if (mcast_sent)
3589 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast_sent);
3590 #ifdef INVARIANTS
3591 	if (iflib_verbose_debug)
3592 		printf("consumed=%d\n", consumed);
3593 #endif
3594 	return (consumed);
3595 }
3596 
3597 static uint32_t
3598 iflib_txq_drain_always(struct ifmp_ring *r)
3599 {
3600 	return (1);
3601 }
3602 
3603 static uint32_t
3604 iflib_txq_drain_free(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx)
3605 {
3606 	int i, avail;
3607 	struct mbuf **mp;
3608 	iflib_txq_t txq;
3609 
3610 	txq = r->cookie;
3611 
3612 	txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3613 	CALLOUT_LOCK(txq);
3614 	callout_stop(&txq->ift_timer);
3615 	CALLOUT_UNLOCK(txq);
3616 
3617 	avail = IDXDIFF(pidx, cidx, r->size);
3618 	for (i = 0; i < avail; i++) {
3619 		mp = _ring_peek_one(r, cidx, i, avail - i);
3620 		if (__predict_false(*mp == (struct mbuf *)txq))
3621 			continue;
3622 		m_freem(*mp);
3623 	}
3624 	MPASS(ifmp_ring_is_stalled(r) == 0);
3625 	return (avail);
3626 }
3627 
3628 static void
3629 iflib_ifmp_purge(iflib_txq_t txq)
3630 {
3631 	struct ifmp_ring *r;
3632 
3633 	r = txq->ift_br;
3634 	r->drain = iflib_txq_drain_free;
3635 	r->can_drain = iflib_txq_drain_always;
3636 
3637 	ifmp_ring_check_drainage(r, r->size);
3638 
3639 	r->drain = iflib_txq_drain;
3640 	r->can_drain = iflib_txq_can_drain;
3641 }
3642 
3643 static void
3644 _task_fn_tx(void *context)
3645 {
3646 	iflib_txq_t txq = context;
3647 	if_ctx_t ctx = txq->ift_ctx;
3648 	struct ifnet *ifp = ctx->ifc_ifp;
3649 	int rc;
3650 
3651 #ifdef IFLIB_DIAGNOSTICS
3652 	txq->ift_cpu_exec_count[curcpu]++;
3653 #endif
3654 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))
3655 		return;
3656 	if (if_getcapenable(ifp) & IFCAP_NETMAP) {
3657 		if (ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false))
3658 			netmap_tx_irq(ifp, txq->ift_id);
3659 		IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id);
3660 		return;
3661 	}
3662 	if (txq->ift_db_pending)
3663 		ifmp_ring_enqueue(txq->ift_br, (void **)&txq, 1, TX_BATCH_SIZE);
3664 	ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE);
3665 	if (ctx->ifc_flags & IFC_LEGACY)
3666 		IFDI_INTR_ENABLE(ctx);
3667 	else {
3668 		rc = IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id);
3669 		KASSERT(rc != ENOTSUP, ("MSI-X support requires queue_intr_enable, but not implemented in driver"));
3670 	}
3671 }
3672 
3673 static void
3674 _task_fn_rx(void *context)
3675 {
3676 	iflib_rxq_t rxq = context;
3677 	if_ctx_t ctx = rxq->ifr_ctx;
3678 	bool more;
3679 	int rc;
3680 	uint16_t budget;
3681 
3682 #ifdef IFLIB_DIAGNOSTICS
3683 	rxq->ifr_cpu_exec_count[curcpu]++;
3684 #endif
3685 	DBG_COUNTER_INC(task_fn_rxs);
3686 	if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)))
3687 		return;
3688 	more = true;
3689 #ifdef DEV_NETMAP
3690 	if (if_getcapenable(ctx->ifc_ifp) & IFCAP_NETMAP) {
3691 		u_int work = 0;
3692 		if (netmap_rx_irq(ctx->ifc_ifp, rxq->ifr_id, &work)) {
3693 			more = false;
3694 		}
3695 	}
3696 #endif
3697 	budget = ctx->ifc_sysctl_rx_budget;
3698 	if (budget == 0)
3699 		budget = 16;	/* XXX */
3700 	if (more == false || (more = iflib_rxeof(rxq, budget)) == false) {
3701 		if (ctx->ifc_flags & IFC_LEGACY)
3702 			IFDI_INTR_ENABLE(ctx);
3703 		else {
3704 			DBG_COUNTER_INC(rx_intr_enables);
3705 			rc = IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id);
3706 			KASSERT(rc != ENOTSUP, ("MSI-X support requires queue_intr_enable, but not implemented in driver"));
3707 		}
3708 	}
3709 	if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)))
3710 		return;
3711 	if (more)
3712 		GROUPTASK_ENQUEUE(&rxq->ifr_task);
3713 }
3714 
3715 static void
3716 _task_fn_admin(void *context)
3717 {
3718 	if_ctx_t ctx = context;
3719 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
3720 	iflib_txq_t txq;
3721 	int i;
3722 	bool oactive, running, do_reset, do_watchdog;
3723 
3724 	STATE_LOCK(ctx);
3725 	running = (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING);
3726 	oactive = (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE);
3727 	do_reset = (ctx->ifc_flags & IFC_DO_RESET);
3728 	do_watchdog = (ctx->ifc_flags & IFC_DO_WATCHDOG);
3729 	ctx->ifc_flags &= ~(IFC_DO_RESET|IFC_DO_WATCHDOG);
3730 	STATE_UNLOCK(ctx);
3731 
3732 	if (!running & !oactive)
3733 		return;
3734 
3735 	CTX_LOCK(ctx);
3736 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) {
3737 		CALLOUT_LOCK(txq);
3738 		callout_stop(&txq->ift_timer);
3739 		CALLOUT_UNLOCK(txq);
3740 	}
3741 	if (do_watchdog) {
3742 		ctx->ifc_watchdog_events++;
3743 		IFDI_WATCHDOG_RESET(ctx);
3744 	}
3745 	IFDI_UPDATE_ADMIN_STATUS(ctx);
3746 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++)
3747 		callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu);
3748 	IFDI_LINK_INTR_ENABLE(ctx);
3749 	if (do_reset)
3750 		iflib_if_init_locked(ctx);
3751 	CTX_UNLOCK(ctx);
3752 
3753 	if (LINK_ACTIVE(ctx) == 0)
3754 		return;
3755 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++)
3756 		iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET);
3757 }
3758 
3759 
3760 static void
3761 _task_fn_iov(void *context)
3762 {
3763 	if_ctx_t ctx = context;
3764 
3765 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))
3766 		return;
3767 
3768 	CTX_LOCK(ctx);
3769 	IFDI_VFLR_HANDLE(ctx);
3770 	CTX_UNLOCK(ctx);
3771 }
3772 
3773 static int
3774 iflib_sysctl_int_delay(SYSCTL_HANDLER_ARGS)
3775 {
3776 	int err;
3777 	if_int_delay_info_t info;
3778 	if_ctx_t ctx;
3779 
3780 	info = (if_int_delay_info_t)arg1;
3781 	ctx = info->iidi_ctx;
3782 	info->iidi_req = req;
3783 	info->iidi_oidp = oidp;
3784 	CTX_LOCK(ctx);
3785 	err = IFDI_SYSCTL_INT_DELAY(ctx, info);
3786 	CTX_UNLOCK(ctx);
3787 	return (err);
3788 }
3789 
3790 /*********************************************************************
3791  *
3792  *  IFNET FUNCTIONS
3793  *
3794  **********************************************************************/
3795 
3796 static void
3797 iflib_if_init_locked(if_ctx_t ctx)
3798 {
3799 	iflib_stop(ctx);
3800 	iflib_init_locked(ctx);
3801 }
3802 
3803 
3804 static void
3805 iflib_if_init(void *arg)
3806 {
3807 	if_ctx_t ctx = arg;
3808 
3809 	CTX_LOCK(ctx);
3810 	iflib_if_init_locked(ctx);
3811 	CTX_UNLOCK(ctx);
3812 }
3813 
3814 static int
3815 iflib_if_transmit(if_t ifp, struct mbuf *m)
3816 {
3817 	if_ctx_t	ctx = if_getsoftc(ifp);
3818 
3819 	iflib_txq_t txq;
3820 	int err, qidx;
3821 
3822 	if (__predict_false((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || !LINK_ACTIVE(ctx))) {
3823 		DBG_COUNTER_INC(tx_frees);
3824 		m_freem(m);
3825 		return (ENOBUFS);
3826 	}
3827 
3828 	MPASS(m->m_nextpkt == NULL);
3829 	qidx = 0;
3830 	if ((NTXQSETS(ctx) > 1) && M_HASHTYPE_GET(m))
3831 		qidx = QIDX(ctx, m);
3832 	/*
3833 	 * XXX calculate buf_ring based on flowid (divvy up bits?)
3834 	 */
3835 	txq = &ctx->ifc_txqs[qidx];
3836 
3837 #ifdef DRIVER_BACKPRESSURE
3838 	if (txq->ift_closed) {
3839 		while (m != NULL) {
3840 			next = m->m_nextpkt;
3841 			m->m_nextpkt = NULL;
3842 			m_freem(m);
3843 			m = next;
3844 		}
3845 		return (ENOBUFS);
3846 	}
3847 #endif
3848 #ifdef notyet
3849 	qidx = count = 0;
3850 	mp = marr;
3851 	next = m;
3852 	do {
3853 		count++;
3854 		next = next->m_nextpkt;
3855 	} while (next != NULL);
3856 
3857 	if (count > nitems(marr))
3858 		if ((mp = malloc(count*sizeof(struct mbuf *), M_IFLIB, M_NOWAIT)) == NULL) {
3859 			/* XXX check nextpkt */
3860 			m_freem(m);
3861 			/* XXX simplify for now */
3862 			DBG_COUNTER_INC(tx_frees);
3863 			return (ENOBUFS);
3864 		}
3865 	for (next = m, i = 0; next != NULL; i++) {
3866 		mp[i] = next;
3867 		next = next->m_nextpkt;
3868 		mp[i]->m_nextpkt = NULL;
3869 	}
3870 #endif
3871 	DBG_COUNTER_INC(tx_seen);
3872 	err = ifmp_ring_enqueue(txq->ift_br, (void **)&m, 1, TX_BATCH_SIZE);
3873 
3874 	GROUPTASK_ENQUEUE(&txq->ift_task);
3875 	if (err) {
3876 		/* support forthcoming later */
3877 #ifdef DRIVER_BACKPRESSURE
3878 		txq->ift_closed = TRUE;
3879 #endif
3880 		ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE);
3881 		m_freem(m);
3882 	}
3883 
3884 	return (err);
3885 }
3886 
3887 static void
3888 iflib_if_qflush(if_t ifp)
3889 {
3890 	if_ctx_t ctx = if_getsoftc(ifp);
3891 	iflib_txq_t txq = ctx->ifc_txqs;
3892 	int i;
3893 
3894 	STATE_LOCK(ctx);
3895 	ctx->ifc_flags |= IFC_QFLUSH;
3896 	STATE_UNLOCK(ctx);
3897 	for (i = 0; i < NTXQSETS(ctx); i++, txq++)
3898 		while (!(ifmp_ring_is_idle(txq->ift_br) || ifmp_ring_is_stalled(txq->ift_br)))
3899 			iflib_txq_check_drain(txq, 0);
3900 	STATE_LOCK(ctx);
3901 	ctx->ifc_flags &= ~IFC_QFLUSH;
3902 	STATE_UNLOCK(ctx);
3903 
3904 	if_qflush(ifp);
3905 }
3906 
3907 
3908 #define IFCAP_FLAGS (IFCAP_TXCSUM_IPV6 | IFCAP_RXCSUM_IPV6 | IFCAP_HWCSUM | IFCAP_LRO | \
3909 		     IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_VLAN_HWTAGGING | IFCAP_HWSTATS | \
3910 		     IFCAP_VLAN_MTU | IFCAP_VLAN_HWFILTER | IFCAP_VLAN_HWTSO)
3911 
3912 static int
3913 iflib_if_ioctl(if_t ifp, u_long command, caddr_t data)
3914 {
3915 	if_ctx_t ctx = if_getsoftc(ifp);
3916 	struct ifreq	*ifr = (struct ifreq *)data;
3917 #if defined(INET) || defined(INET6)
3918 	struct ifaddr	*ifa = (struct ifaddr *)data;
3919 #endif
3920 	bool		avoid_reset = FALSE;
3921 	int		err = 0, reinit = 0, bits;
3922 
3923 	switch (command) {
3924 	case SIOCSIFADDR:
3925 #ifdef INET
3926 		if (ifa->ifa_addr->sa_family == AF_INET)
3927 			avoid_reset = TRUE;
3928 #endif
3929 #ifdef INET6
3930 		if (ifa->ifa_addr->sa_family == AF_INET6)
3931 			avoid_reset = TRUE;
3932 #endif
3933 		/*
3934 		** Calling init results in link renegotiation,
3935 		** so we avoid doing it when possible.
3936 		*/
3937 		if (avoid_reset) {
3938 			if_setflagbits(ifp, IFF_UP,0);
3939 			if (!(if_getdrvflags(ifp)& IFF_DRV_RUNNING))
3940 				reinit = 1;
3941 #ifdef INET
3942 			if (!(if_getflags(ifp) & IFF_NOARP))
3943 				arp_ifinit(ifp, ifa);
3944 #endif
3945 		} else
3946 			err = ether_ioctl(ifp, command, data);
3947 		break;
3948 	case SIOCSIFMTU:
3949 		CTX_LOCK(ctx);
3950 		if (ifr->ifr_mtu == if_getmtu(ifp)) {
3951 			CTX_UNLOCK(ctx);
3952 			break;
3953 		}
3954 		bits = if_getdrvflags(ifp);
3955 		/* stop the driver and free any clusters before proceeding */
3956 		iflib_stop(ctx);
3957 
3958 		if ((err = IFDI_MTU_SET(ctx, ifr->ifr_mtu)) == 0) {
3959 			STATE_LOCK(ctx);
3960 			if (ifr->ifr_mtu > ctx->ifc_max_fl_buf_size)
3961 				ctx->ifc_flags |= IFC_MULTISEG;
3962 			else
3963 				ctx->ifc_flags &= ~IFC_MULTISEG;
3964 			STATE_UNLOCK(ctx);
3965 			err = if_setmtu(ifp, ifr->ifr_mtu);
3966 		}
3967 		iflib_init_locked(ctx);
3968 		STATE_LOCK(ctx);
3969 		if_setdrvflags(ifp, bits);
3970 		STATE_UNLOCK(ctx);
3971 		CTX_UNLOCK(ctx);
3972 		break;
3973 	case SIOCSIFFLAGS:
3974 		CTX_LOCK(ctx);
3975 		if (if_getflags(ifp) & IFF_UP) {
3976 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3977 				if ((if_getflags(ifp) ^ ctx->ifc_if_flags) &
3978 				    (IFF_PROMISC | IFF_ALLMULTI)) {
3979 					err = IFDI_PROMISC_SET(ctx, if_getflags(ifp));
3980 				}
3981 			} else
3982 				reinit = 1;
3983 		} else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3984 			iflib_stop(ctx);
3985 		}
3986 		ctx->ifc_if_flags = if_getflags(ifp);
3987 		CTX_UNLOCK(ctx);
3988 		break;
3989 	case SIOCADDMULTI:
3990 	case SIOCDELMULTI:
3991 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3992 			CTX_LOCK(ctx);
3993 			IFDI_INTR_DISABLE(ctx);
3994 			IFDI_MULTI_SET(ctx);
3995 			IFDI_INTR_ENABLE(ctx);
3996 			CTX_UNLOCK(ctx);
3997 		}
3998 		break;
3999 	case SIOCSIFMEDIA:
4000 		CTX_LOCK(ctx);
4001 		IFDI_MEDIA_SET(ctx);
4002 		CTX_UNLOCK(ctx);
4003 		/* falls thru */
4004 	case SIOCGIFMEDIA:
4005 	case SIOCGIFXMEDIA:
4006 		err = ifmedia_ioctl(ifp, ifr, &ctx->ifc_media, command);
4007 		break;
4008 	case SIOCGI2C:
4009 	{
4010 		struct ifi2creq i2c;
4011 
4012 		err = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
4013 		if (err != 0)
4014 			break;
4015 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
4016 			err = EINVAL;
4017 			break;
4018 		}
4019 		if (i2c.len > sizeof(i2c.data)) {
4020 			err = EINVAL;
4021 			break;
4022 		}
4023 
4024 		if ((err = IFDI_I2C_REQ(ctx, &i2c)) == 0)
4025 			err = copyout(&i2c, ifr_data_get_ptr(ifr),
4026 			    sizeof(i2c));
4027 		break;
4028 	}
4029 	case SIOCSIFCAP:
4030 	{
4031 		int mask, setmask;
4032 
4033 		mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
4034 		setmask = 0;
4035 #ifdef TCP_OFFLOAD
4036 		setmask |= mask & (IFCAP_TOE4|IFCAP_TOE6);
4037 #endif
4038 		setmask |= (mask & IFCAP_FLAGS);
4039 
4040 		if (setmask  & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
4041 			setmask |= (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6);
4042 		if ((mask & IFCAP_WOL) &&
4043 		    (if_getcapabilities(ifp) & IFCAP_WOL) != 0)
4044 			setmask |= (mask & (IFCAP_WOL_MCAST|IFCAP_WOL_MAGIC));
4045 		if_vlancap(ifp);
4046 		/*
4047 		 * want to ensure that traffic has stopped before we change any of the flags
4048 		 */
4049 		if (setmask) {
4050 			CTX_LOCK(ctx);
4051 			bits = if_getdrvflags(ifp);
4052 			if (bits & IFF_DRV_RUNNING)
4053 				iflib_stop(ctx);
4054 			STATE_LOCK(ctx);
4055 			if_togglecapenable(ifp, setmask);
4056 			STATE_UNLOCK(ctx);
4057 			if (bits & IFF_DRV_RUNNING)
4058 				iflib_init_locked(ctx);
4059 			STATE_LOCK(ctx);
4060 			if_setdrvflags(ifp, bits);
4061 			STATE_UNLOCK(ctx);
4062 			CTX_UNLOCK(ctx);
4063 		}
4064 		break;
4065 	    }
4066 	case SIOCGPRIVATE_0:
4067 	case SIOCSDRVSPEC:
4068 	case SIOCGDRVSPEC:
4069 		CTX_LOCK(ctx);
4070 		err = IFDI_PRIV_IOCTL(ctx, command, data);
4071 		CTX_UNLOCK(ctx);
4072 		break;
4073 	default:
4074 		err = ether_ioctl(ifp, command, data);
4075 		break;
4076 	}
4077 	if (reinit)
4078 		iflib_if_init(ctx);
4079 	return (err);
4080 }
4081 
4082 static uint64_t
4083 iflib_if_get_counter(if_t ifp, ift_counter cnt)
4084 {
4085 	if_ctx_t ctx = if_getsoftc(ifp);
4086 
4087 	return (IFDI_GET_COUNTER(ctx, cnt));
4088 }
4089 
4090 /*********************************************************************
4091  *
4092  *  OTHER FUNCTIONS EXPORTED TO THE STACK
4093  *
4094  **********************************************************************/
4095 
4096 static void
4097 iflib_vlan_register(void *arg, if_t ifp, uint16_t vtag)
4098 {
4099 	if_ctx_t ctx = if_getsoftc(ifp);
4100 
4101 	if ((void *)ctx != arg)
4102 		return;
4103 
4104 	if ((vtag == 0) || (vtag > 4095))
4105 		return;
4106 
4107 	CTX_LOCK(ctx);
4108 	IFDI_VLAN_REGISTER(ctx, vtag);
4109 	/* Re-init to load the changes */
4110 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
4111 		iflib_if_init_locked(ctx);
4112 	CTX_UNLOCK(ctx);
4113 }
4114 
4115 static void
4116 iflib_vlan_unregister(void *arg, if_t ifp, uint16_t vtag)
4117 {
4118 	if_ctx_t ctx = if_getsoftc(ifp);
4119 
4120 	if ((void *)ctx != arg)
4121 		return;
4122 
4123 	if ((vtag == 0) || (vtag > 4095))
4124 		return;
4125 
4126 	CTX_LOCK(ctx);
4127 	IFDI_VLAN_UNREGISTER(ctx, vtag);
4128 	/* Re-init to load the changes */
4129 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
4130 		iflib_if_init_locked(ctx);
4131 	CTX_UNLOCK(ctx);
4132 }
4133 
4134 static void
4135 iflib_led_func(void *arg, int onoff)
4136 {
4137 	if_ctx_t ctx = arg;
4138 
4139 	CTX_LOCK(ctx);
4140 	IFDI_LED_FUNC(ctx, onoff);
4141 	CTX_UNLOCK(ctx);
4142 }
4143 
4144 /*********************************************************************
4145  *
4146  *  BUS FUNCTION DEFINITIONS
4147  *
4148  **********************************************************************/
4149 
4150 int
4151 iflib_device_probe(device_t dev)
4152 {
4153 	pci_vendor_info_t *ent;
4154 
4155 	uint16_t	pci_vendor_id, pci_device_id;
4156 	uint16_t	pci_subvendor_id, pci_subdevice_id;
4157 	uint16_t	pci_rev_id;
4158 	if_shared_ctx_t sctx;
4159 
4160 	if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC)
4161 		return (ENOTSUP);
4162 
4163 	pci_vendor_id = pci_get_vendor(dev);
4164 	pci_device_id = pci_get_device(dev);
4165 	pci_subvendor_id = pci_get_subvendor(dev);
4166 	pci_subdevice_id = pci_get_subdevice(dev);
4167 	pci_rev_id = pci_get_revid(dev);
4168 	if (sctx->isc_parse_devinfo != NULL)
4169 		sctx->isc_parse_devinfo(&pci_device_id, &pci_subvendor_id, &pci_subdevice_id, &pci_rev_id);
4170 
4171 	ent = sctx->isc_vendor_info;
4172 	while (ent->pvi_vendor_id != 0) {
4173 		if (pci_vendor_id != ent->pvi_vendor_id) {
4174 			ent++;
4175 			continue;
4176 		}
4177 		if ((pci_device_id == ent->pvi_device_id) &&
4178 		    ((pci_subvendor_id == ent->pvi_subvendor_id) ||
4179 		     (ent->pvi_subvendor_id == 0)) &&
4180 		    ((pci_subdevice_id == ent->pvi_subdevice_id) ||
4181 		     (ent->pvi_subdevice_id == 0)) &&
4182 		    ((pci_rev_id == ent->pvi_rev_id) ||
4183 		     (ent->pvi_rev_id == 0))) {
4184 
4185 			device_set_desc_copy(dev, ent->pvi_name);
4186 			/* this needs to be changed to zero if the bus probing code
4187 			 * ever stops re-probing on best match because the sctx
4188 			 * may have its values over written by register calls
4189 			 * in subsequent probes
4190 			 */
4191 			return (BUS_PROBE_DEFAULT);
4192 		}
4193 		ent++;
4194 	}
4195 	return (ENXIO);
4196 }
4197 
4198 int
4199 iflib_device_register(device_t dev, void *sc, if_shared_ctx_t sctx, if_ctx_t *ctxp)
4200 {
4201 	int err, rid, msix, msix_bar;
4202 	if_ctx_t ctx;
4203 	if_t ifp;
4204 	if_softc_ctx_t scctx;
4205 	int i;
4206 	uint16_t main_txq;
4207 	uint16_t main_rxq;
4208 
4209 
4210 	ctx = malloc(sizeof(* ctx), M_IFLIB, M_WAITOK|M_ZERO);
4211 
4212 	if (sc == NULL) {
4213 		sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO);
4214 		device_set_softc(dev, ctx);
4215 		ctx->ifc_flags |= IFC_SC_ALLOCATED;
4216 	}
4217 
4218 	ctx->ifc_sctx = sctx;
4219 	ctx->ifc_dev = dev;
4220 	ctx->ifc_softc = sc;
4221 
4222 	if ((err = iflib_register(ctx)) != 0) {
4223 		device_printf(dev, "iflib_register failed %d\n", err);
4224 		return (err);
4225 	}
4226 	iflib_add_device_sysctl_pre(ctx);
4227 
4228 	scctx = &ctx->ifc_softc_ctx;
4229 	ifp = ctx->ifc_ifp;
4230 	ctx->ifc_nhwtxqs = sctx->isc_ntxqs;
4231 
4232 	/*
4233 	 * XXX sanity check that ntxd & nrxd are a power of 2
4234 	 */
4235 	if (ctx->ifc_sysctl_ntxqs != 0)
4236 		scctx->isc_ntxqsets = ctx->ifc_sysctl_ntxqs;
4237 	if (ctx->ifc_sysctl_nrxqs != 0)
4238 		scctx->isc_nrxqsets = ctx->ifc_sysctl_nrxqs;
4239 
4240 	for (i = 0; i < sctx->isc_ntxqs; i++) {
4241 		if (ctx->ifc_sysctl_ntxds[i] != 0)
4242 			scctx->isc_ntxd[i] = ctx->ifc_sysctl_ntxds[i];
4243 		else
4244 			scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i];
4245 	}
4246 
4247 	for (i = 0; i < sctx->isc_nrxqs; i++) {
4248 		if (ctx->ifc_sysctl_nrxds[i] != 0)
4249 			scctx->isc_nrxd[i] = ctx->ifc_sysctl_nrxds[i];
4250 		else
4251 			scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i];
4252 	}
4253 
4254 	for (i = 0; i < sctx->isc_nrxqs; i++) {
4255 		if (scctx->isc_nrxd[i] < sctx->isc_nrxd_min[i]) {
4256 			device_printf(dev, "nrxd%d: %d less than nrxd_min %d - resetting to min\n",
4257 				      i, scctx->isc_nrxd[i], sctx->isc_nrxd_min[i]);
4258 			scctx->isc_nrxd[i] = sctx->isc_nrxd_min[i];
4259 		}
4260 		if (scctx->isc_nrxd[i] > sctx->isc_nrxd_max[i]) {
4261 			device_printf(dev, "nrxd%d: %d greater than nrxd_max %d - resetting to max\n",
4262 				      i, scctx->isc_nrxd[i], sctx->isc_nrxd_max[i]);
4263 			scctx->isc_nrxd[i] = sctx->isc_nrxd_max[i];
4264 		}
4265 	}
4266 
4267 	for (i = 0; i < sctx->isc_ntxqs; i++) {
4268 		if (scctx->isc_ntxd[i] < sctx->isc_ntxd_min[i]) {
4269 			device_printf(dev, "ntxd%d: %d less than ntxd_min %d - resetting to min\n",
4270 				      i, scctx->isc_ntxd[i], sctx->isc_ntxd_min[i]);
4271 			scctx->isc_ntxd[i] = sctx->isc_ntxd_min[i];
4272 		}
4273 		if (scctx->isc_ntxd[i] > sctx->isc_ntxd_max[i]) {
4274 			device_printf(dev, "ntxd%d: %d greater than ntxd_max %d - resetting to max\n",
4275 				      i, scctx->isc_ntxd[i], sctx->isc_ntxd_max[i]);
4276 			scctx->isc_ntxd[i] = sctx->isc_ntxd_max[i];
4277 		}
4278 	}
4279 
4280 	if ((err = IFDI_ATTACH_PRE(ctx)) != 0) {
4281 		device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err);
4282 		return (err);
4283 	}
4284 	_iflib_pre_assert(scctx);
4285 	ctx->ifc_txrx = *scctx->isc_txrx;
4286 
4287 #ifdef INVARIANTS
4288 	MPASS(scctx->isc_capenable);
4289 	if (scctx->isc_capenable & IFCAP_TXCSUM)
4290 		MPASS(scctx->isc_tx_csum_flags);
4291 #endif
4292 
4293 	if_setcapabilities(ifp, scctx->isc_capenable | IFCAP_HWSTATS);
4294 	if_setcapenable(ifp, scctx->isc_capenable | IFCAP_HWSTATS);
4295 
4296 	if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets))
4297 		scctx->isc_ntxqsets = scctx->isc_ntxqsets_max;
4298 	if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets))
4299 		scctx->isc_nrxqsets = scctx->isc_nrxqsets_max;
4300 
4301 #ifdef ACPI_DMAR
4302 	if (dmar_get_dma_tag(device_get_parent(dev), dev) != NULL)
4303 		ctx->ifc_flags |= IFC_DMAR;
4304 #elif !(defined(__i386__) || defined(__amd64__))
4305 	/* set unconditionally for !x86 */
4306 	ctx->ifc_flags |= IFC_DMAR;
4307 #endif
4308 
4309 	msix_bar = scctx->isc_msix_bar;
4310 	main_txq = (sctx->isc_flags & IFLIB_HAS_TXCQ) ? 1 : 0;
4311 	main_rxq = (sctx->isc_flags & IFLIB_HAS_RXCQ) ? 1 : 0;
4312 
4313 	/* XXX change for per-queue sizes */
4314 	device_printf(dev, "using %d tx descriptors and %d rx descriptors\n",
4315 		      scctx->isc_ntxd[main_txq], scctx->isc_nrxd[main_rxq]);
4316 	for (i = 0; i < sctx->isc_nrxqs; i++) {
4317 		if (!powerof2(scctx->isc_nrxd[i])) {
4318 			/* round down instead? */
4319 			device_printf(dev, "# rx descriptors must be a power of 2\n");
4320 			err = EINVAL;
4321 			goto fail;
4322 		}
4323 	}
4324 	for (i = 0; i < sctx->isc_ntxqs; i++) {
4325 		if (!powerof2(scctx->isc_ntxd[i])) {
4326 			device_printf(dev,
4327 			    "# tx descriptors must be a power of 2");
4328 			err = EINVAL;
4329 			goto fail;
4330 		}
4331 	}
4332 
4333 	if (scctx->isc_tx_nsegments > scctx->isc_ntxd[main_txq] /
4334 	    MAX_SINGLE_PACKET_FRACTION)
4335 		scctx->isc_tx_nsegments = max(1, scctx->isc_ntxd[main_txq] /
4336 		    MAX_SINGLE_PACKET_FRACTION);
4337 	if (scctx->isc_tx_tso_segments_max > scctx->isc_ntxd[main_txq] /
4338 	    MAX_SINGLE_PACKET_FRACTION)
4339 		scctx->isc_tx_tso_segments_max = max(1,
4340 		    scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION);
4341 
4342 	/*
4343 	 * Protect the stack against modern hardware
4344 	 */
4345 	if (scctx->isc_tx_tso_size_max > FREEBSD_TSO_SIZE_MAX)
4346 		scctx->isc_tx_tso_size_max = FREEBSD_TSO_SIZE_MAX;
4347 
4348 	/* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */
4349 	ifp->if_hw_tsomaxsegcount = scctx->isc_tx_tso_segments_max;
4350 	ifp->if_hw_tsomax = scctx->isc_tx_tso_size_max;
4351 	ifp->if_hw_tsomaxsegsize = scctx->isc_tx_tso_segsize_max;
4352 	if (scctx->isc_rss_table_size == 0)
4353 		scctx->isc_rss_table_size = 64;
4354 	scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1;
4355 
4356 	GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx);
4357 	/* XXX format name */
4358 	taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx, -1, "admin");
4359 
4360 	/* Set up cpu set.  If it fails, use the set of all CPUs. */
4361 	if (bus_get_cpus(dev, INTR_CPUS, sizeof(ctx->ifc_cpus), &ctx->ifc_cpus) != 0) {
4362 		device_printf(dev, "Unable to fetch CPU list\n");
4363 		CPU_COPY(&all_cpus, &ctx->ifc_cpus);
4364 	}
4365 	MPASS(CPU_COUNT(&ctx->ifc_cpus) > 0);
4366 
4367 	/*
4368 	** Now setup MSI or MSI/X, should
4369 	** return us the number of supported
4370 	** vectors. (Will be 1 for MSI)
4371 	*/
4372 	if (sctx->isc_flags & IFLIB_SKIP_MSIX) {
4373 		msix = scctx->isc_vectors;
4374 	} else if (scctx->isc_msix_bar != 0)
4375 	       /*
4376 		* The simple fact that isc_msix_bar is not 0 does not mean we
4377 		* we have a good value there that is known to work.
4378 		*/
4379 		msix = iflib_msix_init(ctx);
4380 	else {
4381 		scctx->isc_vectors = 1;
4382 		scctx->isc_ntxqsets = 1;
4383 		scctx->isc_nrxqsets = 1;
4384 		scctx->isc_intr = IFLIB_INTR_LEGACY;
4385 		msix = 0;
4386 	}
4387 	/* Get memory for the station queues */
4388 	if ((err = iflib_queues_alloc(ctx))) {
4389 		device_printf(dev, "Unable to allocate queue memory\n");
4390 		goto fail;
4391 	}
4392 
4393 	if ((err = iflib_qset_structures_setup(ctx))) {
4394 		device_printf(dev, "qset structure setup failed %d\n", err);
4395 		goto fail_queues;
4396 	}
4397 
4398 	/*
4399 	 * Group taskqueues aren't properly set up until SMP is started,
4400 	 * so we disable interrupts until we can handle them post
4401 	 * SI_SUB_SMP.
4402 	 *
4403 	 * XXX: disabling interrupts doesn't actually work, at least for
4404 	 * the non-MSI case.  When they occur before SI_SUB_SMP completes,
4405 	 * we do null handling and depend on this not causing too large an
4406 	 * interrupt storm.
4407 	 */
4408 	IFDI_INTR_DISABLE(ctx);
4409 	if (msix > 1 && (err = IFDI_MSIX_INTR_ASSIGN(ctx, msix)) != 0) {
4410 		device_printf(dev, "IFDI_MSIX_INTR_ASSIGN failed %d\n", err);
4411 		goto fail_intr_free;
4412 	}
4413 	if (msix <= 1) {
4414 		rid = 0;
4415 		if (scctx->isc_intr == IFLIB_INTR_MSI) {
4416 			MPASS(msix == 1);
4417 			rid = 1;
4418 		}
4419 		if ((err = iflib_legacy_setup(ctx, ctx->isc_legacy_intr, ctx->ifc_softc, &rid, "irq0")) != 0) {
4420 			device_printf(dev, "iflib_legacy_setup failed %d\n", err);
4421 			goto fail_intr_free;
4422 		}
4423 	}
4424 	ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac);
4425 	if ((err = IFDI_ATTACH_POST(ctx)) != 0) {
4426 		device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err);
4427 		goto fail_detach;
4428 	}
4429 	if ((err = iflib_netmap_attach(ctx))) {
4430 		device_printf(ctx->ifc_dev, "netmap attach failed: %d\n", err);
4431 		goto fail_detach;
4432 	}
4433 	*ctxp = ctx;
4434 
4435 	if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter);
4436 	iflib_add_device_sysctl_post(ctx);
4437 	ctx->ifc_flags |= IFC_INIT_DONE;
4438 	return (0);
4439 fail_detach:
4440 	ether_ifdetach(ctx->ifc_ifp);
4441 fail_intr_free:
4442 	if (scctx->isc_intr == IFLIB_INTR_MSIX || scctx->isc_intr == IFLIB_INTR_MSI)
4443 		pci_release_msi(ctx->ifc_dev);
4444 fail_queues:
4445 	/* XXX free queues */
4446 fail:
4447 	IFDI_DETACH(ctx);
4448 	return (err);
4449 }
4450 
4451 int
4452 iflib_device_attach(device_t dev)
4453 {
4454 	if_ctx_t ctx;
4455 	if_shared_ctx_t sctx;
4456 
4457 	if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC)
4458 		return (ENOTSUP);
4459 
4460 	pci_enable_busmaster(dev);
4461 
4462 	return (iflib_device_register(dev, NULL, sctx, &ctx));
4463 }
4464 
4465 int
4466 iflib_device_deregister(if_ctx_t ctx)
4467 {
4468 	if_t ifp = ctx->ifc_ifp;
4469 	iflib_txq_t txq;
4470 	iflib_rxq_t rxq;
4471 	device_t dev = ctx->ifc_dev;
4472 	int i, j;
4473 	struct taskqgroup *tqg;
4474 	iflib_fl_t fl;
4475 
4476 	/* Make sure VLANS are not using driver */
4477 	if (if_vlantrunkinuse(ifp)) {
4478 		device_printf(dev,"Vlan in use, detach first\n");
4479 		return (EBUSY);
4480 	}
4481 
4482 	CTX_LOCK(ctx);
4483 	ctx->ifc_in_detach = 1;
4484 	iflib_stop(ctx);
4485 	CTX_UNLOCK(ctx);
4486 
4487 	/* Unregister VLAN events */
4488 	if (ctx->ifc_vlan_attach_event != NULL)
4489 		EVENTHANDLER_DEREGISTER(vlan_config, ctx->ifc_vlan_attach_event);
4490 	if (ctx->ifc_vlan_detach_event != NULL)
4491 		EVENTHANDLER_DEREGISTER(vlan_unconfig, ctx->ifc_vlan_detach_event);
4492 
4493 	iflib_netmap_detach(ifp);
4494 	ether_ifdetach(ifp);
4495 	/* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/
4496 	CTX_LOCK_DESTROY(ctx);
4497 	if (ctx->ifc_led_dev != NULL)
4498 		led_destroy(ctx->ifc_led_dev);
4499 	/* XXX drain any dependent tasks */
4500 	tqg = qgroup_if_io_tqg;
4501 	for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) {
4502 		callout_drain(&txq->ift_timer);
4503 		if (txq->ift_task.gt_uniq != NULL)
4504 			taskqgroup_detach(tqg, &txq->ift_task);
4505 	}
4506 	for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) {
4507 		if (rxq->ifr_task.gt_uniq != NULL)
4508 			taskqgroup_detach(tqg, &rxq->ifr_task);
4509 
4510 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++)
4511 			free(fl->ifl_rx_bitmap, M_IFLIB);
4512 
4513 	}
4514 	tqg = qgroup_if_config_tqg;
4515 	if (ctx->ifc_admin_task.gt_uniq != NULL)
4516 		taskqgroup_detach(tqg, &ctx->ifc_admin_task);
4517 	if (ctx->ifc_vflr_task.gt_uniq != NULL)
4518 		taskqgroup_detach(tqg, &ctx->ifc_vflr_task);
4519 
4520 	IFDI_DETACH(ctx);
4521 	device_set_softc(ctx->ifc_dev, NULL);
4522 	if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_LEGACY) {
4523 		pci_release_msi(dev);
4524 	}
4525 	if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_MSIX) {
4526 		iflib_irq_free(ctx, &ctx->ifc_legacy_irq);
4527 	}
4528 	if (ctx->ifc_msix_mem != NULL) {
4529 		bus_release_resource(ctx->ifc_dev, SYS_RES_MEMORY,
4530 			ctx->ifc_softc_ctx.isc_msix_bar, ctx->ifc_msix_mem);
4531 		ctx->ifc_msix_mem = NULL;
4532 	}
4533 
4534 	bus_generic_detach(dev);
4535 	if_free(ifp);
4536 
4537 	iflib_tx_structures_free(ctx);
4538 	iflib_rx_structures_free(ctx);
4539 	if (ctx->ifc_flags & IFC_SC_ALLOCATED)
4540 		free(ctx->ifc_softc, M_IFLIB);
4541 	free(ctx, M_IFLIB);
4542 	return (0);
4543 }
4544 
4545 
4546 int
4547 iflib_device_detach(device_t dev)
4548 {
4549 	if_ctx_t ctx = device_get_softc(dev);
4550 
4551 	return (iflib_device_deregister(ctx));
4552 }
4553 
4554 int
4555 iflib_device_suspend(device_t dev)
4556 {
4557 	if_ctx_t ctx = device_get_softc(dev);
4558 
4559 	CTX_LOCK(ctx);
4560 	IFDI_SUSPEND(ctx);
4561 	CTX_UNLOCK(ctx);
4562 
4563 	return bus_generic_suspend(dev);
4564 }
4565 int
4566 iflib_device_shutdown(device_t dev)
4567 {
4568 	if_ctx_t ctx = device_get_softc(dev);
4569 
4570 	CTX_LOCK(ctx);
4571 	IFDI_SHUTDOWN(ctx);
4572 	CTX_UNLOCK(ctx);
4573 
4574 	return bus_generic_suspend(dev);
4575 }
4576 
4577 
4578 int
4579 iflib_device_resume(device_t dev)
4580 {
4581 	if_ctx_t ctx = device_get_softc(dev);
4582 	iflib_txq_t txq = ctx->ifc_txqs;
4583 
4584 	CTX_LOCK(ctx);
4585 	IFDI_RESUME(ctx);
4586 	iflib_init_locked(ctx);
4587 	CTX_UNLOCK(ctx);
4588 	for (int i = 0; i < NTXQSETS(ctx); i++, txq++)
4589 		iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET);
4590 
4591 	return (bus_generic_resume(dev));
4592 }
4593 
4594 int
4595 iflib_device_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *params)
4596 {
4597 	int error;
4598 	if_ctx_t ctx = device_get_softc(dev);
4599 
4600 	CTX_LOCK(ctx);
4601 	error = IFDI_IOV_INIT(ctx, num_vfs, params);
4602 	CTX_UNLOCK(ctx);
4603 
4604 	return (error);
4605 }
4606 
4607 void
4608 iflib_device_iov_uninit(device_t dev)
4609 {
4610 	if_ctx_t ctx = device_get_softc(dev);
4611 
4612 	CTX_LOCK(ctx);
4613 	IFDI_IOV_UNINIT(ctx);
4614 	CTX_UNLOCK(ctx);
4615 }
4616 
4617 int
4618 iflib_device_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *params)
4619 {
4620 	int error;
4621 	if_ctx_t ctx = device_get_softc(dev);
4622 
4623 	CTX_LOCK(ctx);
4624 	error = IFDI_IOV_VF_ADD(ctx, vfnum, params);
4625 	CTX_UNLOCK(ctx);
4626 
4627 	return (error);
4628 }
4629 
4630 /*********************************************************************
4631  *
4632  *  MODULE FUNCTION DEFINITIONS
4633  *
4634  **********************************************************************/
4635 
4636 /*
4637  * - Start a fast taskqueue thread for each core
4638  * - Start a taskqueue for control operations
4639  */
4640 static int
4641 iflib_module_init(void)
4642 {
4643 	return (0);
4644 }
4645 
4646 static int
4647 iflib_module_event_handler(module_t mod, int what, void *arg)
4648 {
4649 	int err;
4650 
4651 	switch (what) {
4652 	case MOD_LOAD:
4653 		if ((err = iflib_module_init()) != 0)
4654 			return (err);
4655 		break;
4656 	case MOD_UNLOAD:
4657 		return (EBUSY);
4658 	default:
4659 		return (EOPNOTSUPP);
4660 	}
4661 
4662 	return (0);
4663 }
4664 
4665 /*********************************************************************
4666  *
4667  *  PUBLIC FUNCTION DEFINITIONS
4668  *     ordered as in iflib.h
4669  *
4670  **********************************************************************/
4671 
4672 
4673 static void
4674 _iflib_assert(if_shared_ctx_t sctx)
4675 {
4676 	MPASS(sctx->isc_tx_maxsize);
4677 	MPASS(sctx->isc_tx_maxsegsize);
4678 
4679 	MPASS(sctx->isc_rx_maxsize);
4680 	MPASS(sctx->isc_rx_nsegments);
4681 	MPASS(sctx->isc_rx_maxsegsize);
4682 
4683 	MPASS(sctx->isc_nrxd_min[0]);
4684 	MPASS(sctx->isc_nrxd_max[0]);
4685 	MPASS(sctx->isc_nrxd_default[0]);
4686 	MPASS(sctx->isc_ntxd_min[0]);
4687 	MPASS(sctx->isc_ntxd_max[0]);
4688 	MPASS(sctx->isc_ntxd_default[0]);
4689 }
4690 
4691 static void
4692 _iflib_pre_assert(if_softc_ctx_t scctx)
4693 {
4694 
4695 	MPASS(scctx->isc_txrx->ift_txd_encap);
4696 	MPASS(scctx->isc_txrx->ift_txd_flush);
4697 	MPASS(scctx->isc_txrx->ift_txd_credits_update);
4698 	MPASS(scctx->isc_txrx->ift_rxd_available);
4699 	MPASS(scctx->isc_txrx->ift_rxd_pkt_get);
4700 	MPASS(scctx->isc_txrx->ift_rxd_refill);
4701 	MPASS(scctx->isc_txrx->ift_rxd_flush);
4702 }
4703 
4704 static int
4705 iflib_register(if_ctx_t ctx)
4706 {
4707 	if_shared_ctx_t sctx = ctx->ifc_sctx;
4708 	driver_t *driver = sctx->isc_driver;
4709 	device_t dev = ctx->ifc_dev;
4710 	if_t ifp;
4711 
4712 	_iflib_assert(sctx);
4713 
4714 	CTX_LOCK_INIT(ctx, device_get_nameunit(ctx->ifc_dev));
4715 
4716 	STATE_LOCK_INIT(ctx, device_get_nameunit(ctx->ifc_dev));
4717 	ifp = ctx->ifc_ifp = if_gethandle(IFT_ETHER);
4718 	if (ifp == NULL) {
4719 		device_printf(dev, "can not allocate ifnet structure\n");
4720 		return (ENOMEM);
4721 	}
4722 
4723 	/*
4724 	 * Initialize our context's device specific methods
4725 	 */
4726 	kobj_init((kobj_t) ctx, (kobj_class_t) driver);
4727 	kobj_class_compile((kobj_class_t) driver);
4728 	driver->refs++;
4729 
4730 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
4731 	if_setsoftc(ifp, ctx);
4732 	if_setdev(ifp, dev);
4733 	if_setinitfn(ifp, iflib_if_init);
4734 	if_setioctlfn(ifp, iflib_if_ioctl);
4735 	if_settransmitfn(ifp, iflib_if_transmit);
4736 	if_setqflushfn(ifp, iflib_if_qflush);
4737 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
4738 
4739 	ctx->ifc_vlan_attach_event =
4740 		EVENTHANDLER_REGISTER(vlan_config, iflib_vlan_register, ctx,
4741 							  EVENTHANDLER_PRI_FIRST);
4742 	ctx->ifc_vlan_detach_event =
4743 		EVENTHANDLER_REGISTER(vlan_unconfig, iflib_vlan_unregister, ctx,
4744 							  EVENTHANDLER_PRI_FIRST);
4745 
4746 	ifmedia_init(&ctx->ifc_media, IFM_IMASK,
4747 					 iflib_media_change, iflib_media_status);
4748 
4749 	return (0);
4750 }
4751 
4752 
4753 static int
4754 iflib_queues_alloc(if_ctx_t ctx)
4755 {
4756 	if_shared_ctx_t sctx = ctx->ifc_sctx;
4757 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
4758 	device_t dev = ctx->ifc_dev;
4759 	int nrxqsets = scctx->isc_nrxqsets;
4760 	int ntxqsets = scctx->isc_ntxqsets;
4761 	iflib_txq_t txq;
4762 	iflib_rxq_t rxq;
4763 	iflib_fl_t fl = NULL;
4764 	int i, j, cpu, err, txconf, rxconf;
4765 	iflib_dma_info_t ifdip;
4766 	uint32_t *rxqsizes = scctx->isc_rxqsizes;
4767 	uint32_t *txqsizes = scctx->isc_txqsizes;
4768 	uint8_t nrxqs = sctx->isc_nrxqs;
4769 	uint8_t ntxqs = sctx->isc_ntxqs;
4770 	int nfree_lists = sctx->isc_nfl ? sctx->isc_nfl : 1;
4771 	caddr_t *vaddrs;
4772 	uint64_t *paddrs;
4773 	struct ifmp_ring **brscp;
4774 
4775 	KASSERT(ntxqs > 0, ("number of queues per qset must be at least 1"));
4776 	KASSERT(nrxqs > 0, ("number of queues per qset must be at least 1"));
4777 
4778 	brscp = NULL;
4779 	txq = NULL;
4780 	rxq = NULL;
4781 
4782 /* Allocate the TX ring struct memory */
4783 	if (!(txq =
4784 	    (iflib_txq_t) malloc(sizeof(struct iflib_txq) *
4785 	    ntxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) {
4786 		device_printf(dev, "Unable to allocate TX ring memory\n");
4787 		err = ENOMEM;
4788 		goto fail;
4789 	}
4790 
4791 	/* Now allocate the RX */
4792 	if (!(rxq =
4793 	    (iflib_rxq_t) malloc(sizeof(struct iflib_rxq) *
4794 	    nrxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) {
4795 		device_printf(dev, "Unable to allocate RX ring memory\n");
4796 		err = ENOMEM;
4797 		goto rx_fail;
4798 	}
4799 
4800 	ctx->ifc_txqs = txq;
4801 	ctx->ifc_rxqs = rxq;
4802 
4803 	/*
4804 	 * XXX handle allocation failure
4805 	 */
4806 	for (txconf = i = 0, cpu = CPU_FIRST(); i < ntxqsets; i++, txconf++, txq++, cpu = CPU_NEXT(cpu)) {
4807 		/* Set up some basics */
4808 
4809 		if ((ifdip = malloc(sizeof(struct iflib_dma_info) * ntxqs, M_IFLIB, M_WAITOK|M_ZERO)) == NULL) {
4810 			device_printf(dev, "failed to allocate iflib_dma_info\n");
4811 			err = ENOMEM;
4812 			goto err_tx_desc;
4813 		}
4814 		txq->ift_ifdi = ifdip;
4815 		for (j = 0; j < ntxqs; j++, ifdip++) {
4816 			if (iflib_dma_alloc(ctx, txqsizes[j], ifdip, BUS_DMA_NOWAIT)) {
4817 				device_printf(dev, "Unable to allocate Descriptor memory\n");
4818 				err = ENOMEM;
4819 				goto err_tx_desc;
4820 			}
4821 			txq->ift_txd_size[j] = scctx->isc_txd_size[j];
4822 			bzero((void *)ifdip->idi_vaddr, txqsizes[j]);
4823 		}
4824 		txq->ift_ctx = ctx;
4825 		txq->ift_id = i;
4826 		if (sctx->isc_flags & IFLIB_HAS_TXCQ) {
4827 			txq->ift_br_offset = 1;
4828 		} else {
4829 			txq->ift_br_offset = 0;
4830 		}
4831 		/* XXX fix this */
4832 		txq->ift_timer.c_cpu = cpu;
4833 
4834 		if (iflib_txsd_alloc(txq)) {
4835 			device_printf(dev, "Critical Failure setting up TX buffers\n");
4836 			err = ENOMEM;
4837 			goto err_tx_desc;
4838 		}
4839 
4840 		/* Initialize the TX lock */
4841 		snprintf(txq->ift_mtx_name, MTX_NAME_LEN, "%s:tx(%d):callout",
4842 		    device_get_nameunit(dev), txq->ift_id);
4843 		mtx_init(&txq->ift_mtx, txq->ift_mtx_name, NULL, MTX_DEF);
4844 		callout_init_mtx(&txq->ift_timer, &txq->ift_mtx, 0);
4845 
4846 		snprintf(txq->ift_db_mtx_name, MTX_NAME_LEN, "%s:tx(%d):db",
4847 			 device_get_nameunit(dev), txq->ift_id);
4848 
4849 		err = ifmp_ring_alloc(&txq->ift_br, 2048, txq, iflib_txq_drain,
4850 				      iflib_txq_can_drain, M_IFLIB, M_WAITOK);
4851 		if (err) {
4852 			/* XXX free any allocated rings */
4853 			device_printf(dev, "Unable to allocate buf_ring\n");
4854 			goto err_tx_desc;
4855 		}
4856 	}
4857 
4858 	for (rxconf = i = 0; i < nrxqsets; i++, rxconf++, rxq++) {
4859 		/* Set up some basics */
4860 
4861 		if ((ifdip = malloc(sizeof(struct iflib_dma_info) * nrxqs, M_IFLIB, M_WAITOK|M_ZERO)) == NULL) {
4862 			device_printf(dev, "failed to allocate iflib_dma_info\n");
4863 			err = ENOMEM;
4864 			goto err_tx_desc;
4865 		}
4866 
4867 		rxq->ifr_ifdi = ifdip;
4868 		/* XXX this needs to be changed if #rx queues != #tx queues */
4869 		rxq->ifr_ntxqirq = 1;
4870 		rxq->ifr_txqid[0] = i;
4871 		for (j = 0; j < nrxqs; j++, ifdip++) {
4872 			if (iflib_dma_alloc(ctx, rxqsizes[j], ifdip, BUS_DMA_NOWAIT)) {
4873 				device_printf(dev, "Unable to allocate Descriptor memory\n");
4874 				err = ENOMEM;
4875 				goto err_tx_desc;
4876 			}
4877 			bzero((void *)ifdip->idi_vaddr, rxqsizes[j]);
4878 		}
4879 		rxq->ifr_ctx = ctx;
4880 		rxq->ifr_id = i;
4881 		if (sctx->isc_flags & IFLIB_HAS_RXCQ) {
4882 			rxq->ifr_fl_offset = 1;
4883 		} else {
4884 			rxq->ifr_fl_offset = 0;
4885 		}
4886 		rxq->ifr_nfl = nfree_lists;
4887 		if (!(fl =
4888 			  (iflib_fl_t) malloc(sizeof(struct iflib_fl) * nfree_lists, M_IFLIB, M_NOWAIT | M_ZERO))) {
4889 			device_printf(dev, "Unable to allocate free list memory\n");
4890 			err = ENOMEM;
4891 			goto err_tx_desc;
4892 		}
4893 		rxq->ifr_fl = fl;
4894 		for (j = 0; j < nfree_lists; j++) {
4895 			fl[j].ifl_rxq = rxq;
4896 			fl[j].ifl_id = j;
4897 			fl[j].ifl_ifdi = &rxq->ifr_ifdi[j + rxq->ifr_fl_offset];
4898 			fl[j].ifl_rxd_size = scctx->isc_rxd_size[j];
4899 		}
4900         /* Allocate receive buffers for the ring*/
4901 		if (iflib_rxsd_alloc(rxq)) {
4902 			device_printf(dev,
4903 			    "Critical Failure setting up receive buffers\n");
4904 			err = ENOMEM;
4905 			goto err_rx_desc;
4906 		}
4907 
4908 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++)
4909 			fl->ifl_rx_bitmap = bit_alloc(fl->ifl_size, M_IFLIB, M_WAITOK|M_ZERO);
4910 	}
4911 
4912 	/* TXQs */
4913 	vaddrs = malloc(sizeof(caddr_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK);
4914 	paddrs = malloc(sizeof(uint64_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK);
4915 	for (i = 0; i < ntxqsets; i++) {
4916 		iflib_dma_info_t di = ctx->ifc_txqs[i].ift_ifdi;
4917 
4918 		for (j = 0; j < ntxqs; j++, di++) {
4919 			vaddrs[i*ntxqs + j] = di->idi_vaddr;
4920 			paddrs[i*ntxqs + j] = di->idi_paddr;
4921 		}
4922 	}
4923 	if ((err = IFDI_TX_QUEUES_ALLOC(ctx, vaddrs, paddrs, ntxqs, ntxqsets)) != 0) {
4924 		device_printf(ctx->ifc_dev, "device queue allocation failed\n");
4925 		iflib_tx_structures_free(ctx);
4926 		free(vaddrs, M_IFLIB);
4927 		free(paddrs, M_IFLIB);
4928 		goto err_rx_desc;
4929 	}
4930 	free(vaddrs, M_IFLIB);
4931 	free(paddrs, M_IFLIB);
4932 
4933 	/* RXQs */
4934 	vaddrs = malloc(sizeof(caddr_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK);
4935 	paddrs = malloc(sizeof(uint64_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK);
4936 	for (i = 0; i < nrxqsets; i++) {
4937 		iflib_dma_info_t di = ctx->ifc_rxqs[i].ifr_ifdi;
4938 
4939 		for (j = 0; j < nrxqs; j++, di++) {
4940 			vaddrs[i*nrxqs + j] = di->idi_vaddr;
4941 			paddrs[i*nrxqs + j] = di->idi_paddr;
4942 		}
4943 	}
4944 	if ((err = IFDI_RX_QUEUES_ALLOC(ctx, vaddrs, paddrs, nrxqs, nrxqsets)) != 0) {
4945 		device_printf(ctx->ifc_dev, "device queue allocation failed\n");
4946 		iflib_tx_structures_free(ctx);
4947 		free(vaddrs, M_IFLIB);
4948 		free(paddrs, M_IFLIB);
4949 		goto err_rx_desc;
4950 	}
4951 	free(vaddrs, M_IFLIB);
4952 	free(paddrs, M_IFLIB);
4953 
4954 	return (0);
4955 
4956 /* XXX handle allocation failure changes */
4957 err_rx_desc:
4958 err_tx_desc:
4959 	if (ctx->ifc_rxqs != NULL)
4960 		free(ctx->ifc_rxqs, M_IFLIB);
4961 	ctx->ifc_rxqs = NULL;
4962 	if (ctx->ifc_txqs != NULL)
4963 		free(ctx->ifc_txqs, M_IFLIB);
4964 	ctx->ifc_txqs = NULL;
4965 rx_fail:
4966 	if (brscp != NULL)
4967 		free(brscp, M_IFLIB);
4968 	if (rxq != NULL)
4969 		free(rxq, M_IFLIB);
4970 	if (txq != NULL)
4971 		free(txq, M_IFLIB);
4972 fail:
4973 	return (err);
4974 }
4975 
4976 static int
4977 iflib_tx_structures_setup(if_ctx_t ctx)
4978 {
4979 	iflib_txq_t txq = ctx->ifc_txqs;
4980 	int i;
4981 
4982 	for (i = 0; i < NTXQSETS(ctx); i++, txq++)
4983 		iflib_txq_setup(txq);
4984 
4985 	return (0);
4986 }
4987 
4988 static void
4989 iflib_tx_structures_free(if_ctx_t ctx)
4990 {
4991 	iflib_txq_t txq = ctx->ifc_txqs;
4992 	int i, j;
4993 
4994 	for (i = 0; i < NTXQSETS(ctx); i++, txq++) {
4995 		iflib_txq_destroy(txq);
4996 		for (j = 0; j < ctx->ifc_nhwtxqs; j++)
4997 			iflib_dma_free(&txq->ift_ifdi[j]);
4998 	}
4999 	free(ctx->ifc_txqs, M_IFLIB);
5000 	ctx->ifc_txqs = NULL;
5001 	IFDI_QUEUES_FREE(ctx);
5002 }
5003 
5004 /*********************************************************************
5005  *
5006  *  Initialize all receive rings.
5007  *
5008  **********************************************************************/
5009 static int
5010 iflib_rx_structures_setup(if_ctx_t ctx)
5011 {
5012 	iflib_rxq_t rxq = ctx->ifc_rxqs;
5013 	int q;
5014 #if defined(INET6) || defined(INET)
5015 	int i, err;
5016 #endif
5017 
5018 	for (q = 0; q < ctx->ifc_softc_ctx.isc_nrxqsets; q++, rxq++) {
5019 #if defined(INET6) || defined(INET)
5020 		tcp_lro_free(&rxq->ifr_lc);
5021 		if ((err = tcp_lro_init_args(&rxq->ifr_lc, ctx->ifc_ifp,
5022 		    TCP_LRO_ENTRIES, min(1024,
5023 		    ctx->ifc_softc_ctx.isc_nrxd[rxq->ifr_fl_offset]))) != 0) {
5024 			device_printf(ctx->ifc_dev, "LRO Initialization failed!\n");
5025 			goto fail;
5026 		}
5027 		rxq->ifr_lro_enabled = TRUE;
5028 #endif
5029 		IFDI_RXQ_SETUP(ctx, rxq->ifr_id);
5030 	}
5031 	return (0);
5032 #if defined(INET6) || defined(INET)
5033 fail:
5034 	/*
5035 	 * Free RX software descriptors allocated so far, we will only handle
5036 	 * the rings that completed, the failing case will have
5037 	 * cleaned up for itself. 'q' failed, so its the terminus.
5038 	 */
5039 	rxq = ctx->ifc_rxqs;
5040 	for (i = 0; i < q; ++i, rxq++) {
5041 		iflib_rx_sds_free(rxq);
5042 		rxq->ifr_cq_gen = rxq->ifr_cq_cidx = rxq->ifr_cq_pidx = 0;
5043 	}
5044 	return (err);
5045 #endif
5046 }
5047 
5048 /*********************************************************************
5049  *
5050  *  Free all receive rings.
5051  *
5052  **********************************************************************/
5053 static void
5054 iflib_rx_structures_free(if_ctx_t ctx)
5055 {
5056 	iflib_rxq_t rxq = ctx->ifc_rxqs;
5057 
5058 	for (int i = 0; i < ctx->ifc_softc_ctx.isc_nrxqsets; i++, rxq++) {
5059 		iflib_rx_sds_free(rxq);
5060 	}
5061 }
5062 
5063 static int
5064 iflib_qset_structures_setup(if_ctx_t ctx)
5065 {
5066 	int err;
5067 
5068 	if ((err = iflib_tx_structures_setup(ctx)) != 0)
5069 		return (err);
5070 
5071 	if ((err = iflib_rx_structures_setup(ctx)) != 0) {
5072 		device_printf(ctx->ifc_dev, "iflib_rx_structures_setup failed: %d\n", err);
5073 		iflib_tx_structures_free(ctx);
5074 		iflib_rx_structures_free(ctx);
5075 	}
5076 	return (err);
5077 }
5078 
5079 int
5080 iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid,
5081 				driver_filter_t filter, void *filter_arg, driver_intr_t handler, void *arg, char *name)
5082 {
5083 
5084 	return (_iflib_irq_alloc(ctx, irq, rid, filter, handler, arg, name));
5085 }
5086 
5087 #ifdef SMP
5088 static int
5089 find_nth(if_ctx_t ctx, int qid)
5090 {
5091 	cpuset_t cpus;
5092 	int i, cpuid, eqid, count;
5093 
5094 	CPU_COPY(&ctx->ifc_cpus, &cpus);
5095 	count = CPU_COUNT(&cpus);
5096 	eqid = qid % count;
5097 	/* clear up to the qid'th bit */
5098 	for (i = 0; i < eqid; i++) {
5099 		cpuid = CPU_FFS(&cpus);
5100 		MPASS(cpuid != 0);
5101 		CPU_CLR(cpuid-1, &cpus);
5102 	}
5103 	cpuid = CPU_FFS(&cpus);
5104 	MPASS(cpuid != 0);
5105 	return (cpuid-1);
5106 }
5107 
5108 #ifdef SCHED_ULE
5109 extern struct cpu_group *cpu_top;              /* CPU topology */
5110 
5111 static int
5112 find_child_with_core(int cpu, struct cpu_group *grp)
5113 {
5114 	int i;
5115 
5116 	if (grp->cg_children == 0)
5117 		return -1;
5118 
5119 	MPASS(grp->cg_child);
5120 	for (i = 0; i < grp->cg_children; i++) {
5121 		if (CPU_ISSET(cpu, &grp->cg_child[i].cg_mask))
5122 			return i;
5123 	}
5124 
5125 	return -1;
5126 }
5127 
5128 /*
5129  * Find the nth "close" core to the specified core
5130  * "close" is defined as the deepest level that shares
5131  * at least an L2 cache.  With threads, this will be
5132  * threads on the same core.  If the sahred cache is L3
5133  * or higher, simply returns the same core.
5134  */
5135 static int
5136 find_close_core(int cpu, int core_offset)
5137 {
5138 	struct cpu_group *grp;
5139 	int i;
5140 	int fcpu;
5141 	cpuset_t cs;
5142 
5143 	grp = cpu_top;
5144 	if (grp == NULL)
5145 		return cpu;
5146 	i = 0;
5147 	while ((i = find_child_with_core(cpu, grp)) != -1) {
5148 		/* If the child only has one cpu, don't descend */
5149 		if (grp->cg_child[i].cg_count <= 1)
5150 			break;
5151 		grp = &grp->cg_child[i];
5152 	}
5153 
5154 	/* If they don't share at least an L2 cache, use the same CPU */
5155 	if (grp->cg_level > CG_SHARE_L2 || grp->cg_level == CG_SHARE_NONE)
5156 		return cpu;
5157 
5158 	/* Now pick one */
5159 	CPU_COPY(&grp->cg_mask, &cs);
5160 
5161 	/* Add the selected CPU offset to core offset. */
5162 	for (i = 0; (fcpu = CPU_FFS(&cs)) != 0; i++) {
5163 		if (fcpu - 1 == cpu)
5164 			break;
5165 		CPU_CLR(fcpu - 1, &cs);
5166 	}
5167 	MPASS(fcpu);
5168 
5169 	core_offset += i;
5170 
5171 	CPU_COPY(&grp->cg_mask, &cs);
5172 	for (i = core_offset % grp->cg_count; i > 0; i--) {
5173 		MPASS(CPU_FFS(&cs));
5174 		CPU_CLR(CPU_FFS(&cs) - 1, &cs);
5175 	}
5176 	MPASS(CPU_FFS(&cs));
5177 	return CPU_FFS(&cs) - 1;
5178 }
5179 #else
5180 static int
5181 find_close_core(int cpu, int core_offset __unused)
5182 {
5183 	return cpu;
5184 }
5185 #endif
5186 
5187 static int
5188 get_core_offset(if_ctx_t ctx, iflib_intr_type_t type, int qid)
5189 {
5190 	switch (type) {
5191 	case IFLIB_INTR_TX:
5192 		/* TX queues get cores which share at least an L2 cache with the corresponding RX queue */
5193 		/* XXX handle multiple RX threads per core and more than two core per L2 group */
5194 		return qid / CPU_COUNT(&ctx->ifc_cpus) + 1;
5195 	case IFLIB_INTR_RX:
5196 	case IFLIB_INTR_RXTX:
5197 		/* RX queues get the specified core */
5198 		return qid / CPU_COUNT(&ctx->ifc_cpus);
5199 	default:
5200 		return -1;
5201 	}
5202 }
5203 #else
5204 #define get_core_offset(ctx, type, qid)	CPU_FIRST()
5205 #define find_close_core(cpuid, tid)	CPU_FIRST()
5206 #define find_nth(ctx, gid)		CPU_FIRST()
5207 #endif
5208 
5209 /* Just to avoid copy/paste */
5210 static inline int
5211 iflib_irq_set_affinity(if_ctx_t ctx, int irq, iflib_intr_type_t type, int qid,
5212     struct grouptask *gtask, struct taskqgroup *tqg, void *uniq, char *name)
5213 {
5214 	int cpuid;
5215 	int err, tid;
5216 
5217 	cpuid = find_nth(ctx, qid);
5218 	tid = get_core_offset(ctx, type, qid);
5219 	MPASS(tid >= 0);
5220 	cpuid = find_close_core(cpuid, tid);
5221 	err = taskqgroup_attach_cpu(tqg, gtask, uniq, cpuid, irq, name);
5222 	if (err) {
5223 		device_printf(ctx->ifc_dev, "taskqgroup_attach_cpu failed %d\n", err);
5224 		return (err);
5225 	}
5226 #ifdef notyet
5227 	if (cpuid > ctx->ifc_cpuid_highest)
5228 		ctx->ifc_cpuid_highest = cpuid;
5229 #endif
5230 	return 0;
5231 }
5232 
5233 int
5234 iflib_irq_alloc_generic(if_ctx_t ctx, if_irq_t irq, int rid,
5235 						iflib_intr_type_t type, driver_filter_t *filter,
5236 						void *filter_arg, int qid, char *name)
5237 {
5238 	struct grouptask *gtask;
5239 	struct taskqgroup *tqg;
5240 	iflib_filter_info_t info;
5241 	gtask_fn_t *fn;
5242 	int tqrid, err;
5243 	driver_filter_t *intr_fast;
5244 	void *q;
5245 
5246 	info = &ctx->ifc_filter_info;
5247 	tqrid = rid;
5248 
5249 	switch (type) {
5250 	/* XXX merge tx/rx for netmap? */
5251 	case IFLIB_INTR_TX:
5252 		q = &ctx->ifc_txqs[qid];
5253 		info = &ctx->ifc_txqs[qid].ift_filter_info;
5254 		gtask = &ctx->ifc_txqs[qid].ift_task;
5255 		tqg = qgroup_if_io_tqg;
5256 		fn = _task_fn_tx;
5257 		intr_fast = iflib_fast_intr;
5258 		GROUPTASK_INIT(gtask, 0, fn, q);
5259 		break;
5260 	case IFLIB_INTR_RX:
5261 		q = &ctx->ifc_rxqs[qid];
5262 		info = &ctx->ifc_rxqs[qid].ifr_filter_info;
5263 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
5264 		tqg = qgroup_if_io_tqg;
5265 		fn = _task_fn_rx;
5266 		intr_fast = iflib_fast_intr;
5267 		GROUPTASK_INIT(gtask, 0, fn, q);
5268 		break;
5269 	case IFLIB_INTR_RXTX:
5270 		q = &ctx->ifc_rxqs[qid];
5271 		info = &ctx->ifc_rxqs[qid].ifr_filter_info;
5272 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
5273 		tqg = qgroup_if_io_tqg;
5274 		fn = _task_fn_rx;
5275 		intr_fast = iflib_fast_intr_rxtx;
5276 		GROUPTASK_INIT(gtask, 0, fn, q);
5277 		break;
5278 	case IFLIB_INTR_ADMIN:
5279 		q = ctx;
5280 		tqrid = -1;
5281 		info = &ctx->ifc_filter_info;
5282 		gtask = &ctx->ifc_admin_task;
5283 		tqg = qgroup_if_config_tqg;
5284 		fn = _task_fn_admin;
5285 		intr_fast = iflib_fast_intr_ctx;
5286 		break;
5287 	default:
5288 		panic("unknown net intr type");
5289 	}
5290 
5291 	info->ifi_filter = filter;
5292 	info->ifi_filter_arg = filter_arg;
5293 	info->ifi_task = gtask;
5294 	info->ifi_ctx = q;
5295 
5296 	err = _iflib_irq_alloc(ctx, irq, rid, intr_fast, NULL, info,  name);
5297 	if (err != 0) {
5298 		device_printf(ctx->ifc_dev, "_iflib_irq_alloc failed %d\n", err);
5299 		return (err);
5300 	}
5301 	if (type == IFLIB_INTR_ADMIN)
5302 		return (0);
5303 
5304 	if (tqrid != -1) {
5305 		err = iflib_irq_set_affinity(ctx, rman_get_start(irq->ii_res), type, qid, gtask, tqg, q, name);
5306 		if (err)
5307 			return (err);
5308 	} else {
5309 		taskqgroup_attach(tqg, gtask, q, rman_get_start(irq->ii_res), name);
5310 	}
5311 
5312 	return (0);
5313 }
5314 
5315 void
5316 iflib_softirq_alloc_generic(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type,  void *arg, int qid, char *name)
5317 {
5318 	struct grouptask *gtask;
5319 	struct taskqgroup *tqg;
5320 	gtask_fn_t *fn;
5321 	void *q;
5322 	int irq_num = -1;
5323 	int err;
5324 
5325 	switch (type) {
5326 	case IFLIB_INTR_TX:
5327 		q = &ctx->ifc_txqs[qid];
5328 		gtask = &ctx->ifc_txqs[qid].ift_task;
5329 		tqg = qgroup_if_io_tqg;
5330 		fn = _task_fn_tx;
5331 		if (irq != NULL)
5332 			irq_num = rman_get_start(irq->ii_res);
5333 		break;
5334 	case IFLIB_INTR_RX:
5335 		q = &ctx->ifc_rxqs[qid];
5336 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
5337 		tqg = qgroup_if_io_tqg;
5338 		fn = _task_fn_rx;
5339 		if (irq != NULL)
5340 			irq_num = rman_get_start(irq->ii_res);
5341 		break;
5342 	case IFLIB_INTR_IOV:
5343 		q = ctx;
5344 		gtask = &ctx->ifc_vflr_task;
5345 		tqg = qgroup_if_config_tqg;
5346 		fn = _task_fn_iov;
5347 		break;
5348 	default:
5349 		panic("unknown net intr type");
5350 	}
5351 	GROUPTASK_INIT(gtask, 0, fn, q);
5352 	if (irq_num != -1) {
5353 		err = iflib_irq_set_affinity(ctx, irq_num, type, qid, gtask, tqg, q, name);
5354 		if (err)
5355 			taskqgroup_attach(tqg, gtask, q, irq_num, name);
5356 	}
5357 	else {
5358 		taskqgroup_attach(tqg, gtask, q, irq_num, name);
5359 	}
5360 }
5361 
5362 void
5363 iflib_irq_free(if_ctx_t ctx, if_irq_t irq)
5364 {
5365 	if (irq->ii_tag)
5366 		bus_teardown_intr(ctx->ifc_dev, irq->ii_res, irq->ii_tag);
5367 
5368 	if (irq->ii_res)
5369 		bus_release_resource(ctx->ifc_dev, SYS_RES_IRQ, irq->ii_rid, irq->ii_res);
5370 }
5371 
5372 static int
5373 iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filter_arg, int *rid, char *name)
5374 {
5375 	iflib_txq_t txq = ctx->ifc_txqs;
5376 	iflib_rxq_t rxq = ctx->ifc_rxqs;
5377 	if_irq_t irq = &ctx->ifc_legacy_irq;
5378 	iflib_filter_info_t info;
5379 	struct grouptask *gtask;
5380 	struct taskqgroup *tqg;
5381 	gtask_fn_t *fn;
5382 	int tqrid;
5383 	void *q;
5384 	int err;
5385 
5386 	q = &ctx->ifc_rxqs[0];
5387 	info = &rxq[0].ifr_filter_info;
5388 	gtask = &rxq[0].ifr_task;
5389 	tqg = qgroup_if_io_tqg;
5390 	tqrid = irq->ii_rid = *rid;
5391 	fn = _task_fn_rx;
5392 
5393 	ctx->ifc_flags |= IFC_LEGACY;
5394 	info->ifi_filter = filter;
5395 	info->ifi_filter_arg = filter_arg;
5396 	info->ifi_task = gtask;
5397 	info->ifi_ctx = ctx;
5398 
5399 	/* We allocate a single interrupt resource */
5400 	if ((err = _iflib_irq_alloc(ctx, irq, tqrid, iflib_fast_intr_ctx, NULL, info, name)) != 0)
5401 		return (err);
5402 	GROUPTASK_INIT(gtask, 0, fn, q);
5403 	taskqgroup_attach(tqg, gtask, q, rman_get_start(irq->ii_res), name);
5404 
5405 	GROUPTASK_INIT(&txq->ift_task, 0, _task_fn_tx, txq);
5406 	taskqgroup_attach(qgroup_if_io_tqg, &txq->ift_task, txq, rman_get_start(irq->ii_res), "tx");
5407 	return (0);
5408 }
5409 
5410 void
5411 iflib_led_create(if_ctx_t ctx)
5412 {
5413 
5414 	ctx->ifc_led_dev = led_create(iflib_led_func, ctx,
5415 	    device_get_nameunit(ctx->ifc_dev));
5416 }
5417 
5418 void
5419 iflib_tx_intr_deferred(if_ctx_t ctx, int txqid)
5420 {
5421 
5422 	GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task);
5423 }
5424 
5425 void
5426 iflib_rx_intr_deferred(if_ctx_t ctx, int rxqid)
5427 {
5428 
5429 	GROUPTASK_ENQUEUE(&ctx->ifc_rxqs[rxqid].ifr_task);
5430 }
5431 
5432 void
5433 iflib_admin_intr_deferred(if_ctx_t ctx)
5434 {
5435 #ifdef INVARIANTS
5436 	struct grouptask *gtask;
5437 
5438 	gtask = &ctx->ifc_admin_task;
5439 	MPASS(gtask != NULL && gtask->gt_taskqueue != NULL);
5440 #endif
5441 
5442 	GROUPTASK_ENQUEUE(&ctx->ifc_admin_task);
5443 }
5444 
5445 void
5446 iflib_iov_intr_deferred(if_ctx_t ctx)
5447 {
5448 
5449 	GROUPTASK_ENQUEUE(&ctx->ifc_vflr_task);
5450 }
5451 
5452 void
5453 iflib_io_tqg_attach(struct grouptask *gt, void *uniq, int cpu, char *name)
5454 {
5455 
5456 	taskqgroup_attach_cpu(qgroup_if_io_tqg, gt, uniq, cpu, -1, name);
5457 }
5458 
5459 void
5460 iflib_config_gtask_init(if_ctx_t ctx, struct grouptask *gtask, gtask_fn_t *fn,
5461 	char *name)
5462 {
5463 
5464 	GROUPTASK_INIT(gtask, 0, fn, ctx);
5465 	taskqgroup_attach(qgroup_if_config_tqg, gtask, gtask, -1, name);
5466 }
5467 
5468 void
5469 iflib_config_gtask_deinit(struct grouptask *gtask)
5470 {
5471 
5472 	taskqgroup_detach(qgroup_if_config_tqg, gtask);
5473 }
5474 
5475 void
5476 iflib_link_state_change(if_ctx_t ctx, int link_state, uint64_t baudrate)
5477 {
5478 	if_t ifp = ctx->ifc_ifp;
5479 	iflib_txq_t txq = ctx->ifc_txqs;
5480 
5481 	if_setbaudrate(ifp, baudrate);
5482 	if (baudrate >= IF_Gbps(10)) {
5483 		STATE_LOCK(ctx);
5484 		ctx->ifc_flags |= IFC_PREFETCH;
5485 		STATE_UNLOCK(ctx);
5486 	}
5487 	/* If link down, disable watchdog */
5488 	if ((ctx->ifc_link_state == LINK_STATE_UP) && (link_state == LINK_STATE_DOWN)) {
5489 		for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxqsets; i++, txq++)
5490 			txq->ift_qstatus = IFLIB_QUEUE_IDLE;
5491 	}
5492 	ctx->ifc_link_state = link_state;
5493 	if_link_state_change(ifp, link_state);
5494 }
5495 
5496 static int
5497 iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq)
5498 {
5499 	int credits;
5500 #ifdef INVARIANTS
5501 	int credits_pre = txq->ift_cidx_processed;
5502 #endif
5503 
5504 	if (ctx->isc_txd_credits_update == NULL)
5505 		return (0);
5506 
5507 	if ((credits = ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, true)) == 0)
5508 		return (0);
5509 
5510 	txq->ift_processed += credits;
5511 	txq->ift_cidx_processed += credits;
5512 
5513 	MPASS(credits_pre + credits == txq->ift_cidx_processed);
5514 	if (txq->ift_cidx_processed >= txq->ift_size)
5515 		txq->ift_cidx_processed -= txq->ift_size;
5516 	return (credits);
5517 }
5518 
5519 static int
5520 iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget)
5521 {
5522 
5523 	return (ctx->isc_rxd_available(ctx->ifc_softc, rxq->ifr_id, cidx,
5524 	    budget));
5525 }
5526 
5527 void
5528 iflib_add_int_delay_sysctl(if_ctx_t ctx, const char *name,
5529 	const char *description, if_int_delay_info_t info,
5530 	int offset, int value)
5531 {
5532 	info->iidi_ctx = ctx;
5533 	info->iidi_offset = offset;
5534 	info->iidi_value = value;
5535 	SYSCTL_ADD_PROC(device_get_sysctl_ctx(ctx->ifc_dev),
5536 	    SYSCTL_CHILDREN(device_get_sysctl_tree(ctx->ifc_dev)),
5537 	    OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW,
5538 	    info, 0, iflib_sysctl_int_delay, "I", description);
5539 }
5540 
5541 struct mtx *
5542 iflib_ctx_lock_get(if_ctx_t ctx)
5543 {
5544 
5545 	return (&ctx->ifc_ctx_mtx);
5546 }
5547 
5548 static int
5549 iflib_msix_init(if_ctx_t ctx)
5550 {
5551 	device_t dev = ctx->ifc_dev;
5552 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5553 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
5554 	int vectors, queues, rx_queues, tx_queues, queuemsgs, msgs;
5555 	int iflib_num_tx_queues, iflib_num_rx_queues;
5556 	int err, admincnt, bar;
5557 
5558 	iflib_num_tx_queues = ctx->ifc_sysctl_ntxqs;
5559 	iflib_num_rx_queues = ctx->ifc_sysctl_nrxqs;
5560 
5561 	device_printf(dev, "msix_init qsets capped at %d\n", imax(scctx->isc_ntxqsets, scctx->isc_nrxqsets));
5562 
5563 	bar = ctx->ifc_softc_ctx.isc_msix_bar;
5564 	admincnt = sctx->isc_admin_intrcnt;
5565 	/* Override by global tuneable */
5566 	{
5567 		int i;
5568 		size_t len = sizeof(i);
5569 		err = kernel_sysctlbyname(curthread, "hw.pci.enable_msix", &i, &len, NULL, 0, NULL, 0);
5570 		if (err == 0) {
5571 			if (i == 0)
5572 				goto msi;
5573 		}
5574 		else {
5575 			device_printf(dev, "unable to read hw.pci.enable_msix.");
5576 		}
5577 	}
5578 	/* Override by tuneable */
5579 	if (scctx->isc_disable_msix)
5580 		goto msi;
5581 
5582 	/*
5583 	** When used in a virtualized environment
5584 	** PCI BUSMASTER capability may not be set
5585 	** so explicity set it here and rewrite
5586 	** the ENABLE in the MSIX control register
5587 	** at this point to cause the host to
5588 	** successfully initialize us.
5589 	*/
5590 	{
5591 		int msix_ctrl, rid;
5592 
5593  		pci_enable_busmaster(dev);
5594 		rid = 0;
5595 		if (pci_find_cap(dev, PCIY_MSIX, &rid) == 0 && rid != 0) {
5596 			rid += PCIR_MSIX_CTRL;
5597 			msix_ctrl = pci_read_config(dev, rid, 2);
5598 			msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
5599 			pci_write_config(dev, rid, msix_ctrl, 2);
5600 		} else {
5601 			device_printf(dev, "PCIY_MSIX capability not found; "
5602 			                   "or rid %d == 0.\n", rid);
5603 			goto msi;
5604 		}
5605 	}
5606 
5607 	/*
5608 	 * bar == -1 => "trust me I know what I'm doing"
5609 	 * Some drivers are for hardware that is so shoddily
5610 	 * documented that no one knows which bars are which
5611 	 * so the developer has to map all bars. This hack
5612 	 * allows shoddy garbage to use msix in this framework.
5613 	 */
5614 	if (bar != -1) {
5615 		ctx->ifc_msix_mem = bus_alloc_resource_any(dev,
5616 	            SYS_RES_MEMORY, &bar, RF_ACTIVE);
5617 		if (ctx->ifc_msix_mem == NULL) {
5618 			/* May not be enabled */
5619 			device_printf(dev, "Unable to map MSIX table \n");
5620 			goto msi;
5621 		}
5622 	}
5623 	/* First try MSI/X */
5624 	if ((msgs = pci_msix_count(dev)) == 0) { /* system has msix disabled */
5625 		device_printf(dev, "System has MSIX disabled \n");
5626 		bus_release_resource(dev, SYS_RES_MEMORY,
5627 		    bar, ctx->ifc_msix_mem);
5628 		ctx->ifc_msix_mem = NULL;
5629 		goto msi;
5630 	}
5631 #if IFLIB_DEBUG
5632 	/* use only 1 qset in debug mode */
5633 	queuemsgs = min(msgs - admincnt, 1);
5634 #else
5635 	queuemsgs = msgs - admincnt;
5636 #endif
5637 #ifdef RSS
5638 	queues = imin(queuemsgs, rss_getnumbuckets());
5639 #else
5640 	queues = queuemsgs;
5641 #endif
5642 	queues = imin(CPU_COUNT(&ctx->ifc_cpus), queues);
5643 	device_printf(dev, "pxm cpus: %d queue msgs: %d admincnt: %d\n",
5644 				  CPU_COUNT(&ctx->ifc_cpus), queuemsgs, admincnt);
5645 #ifdef  RSS
5646 	/* If we're doing RSS, clamp at the number of RSS buckets */
5647 	if (queues > rss_getnumbuckets())
5648 		queues = rss_getnumbuckets();
5649 #endif
5650 	if (iflib_num_rx_queues > 0 && iflib_num_rx_queues < queuemsgs - admincnt)
5651 		rx_queues = iflib_num_rx_queues;
5652 	else
5653 		rx_queues = queues;
5654 
5655 	if (rx_queues > scctx->isc_nrxqsets)
5656 		rx_queues = scctx->isc_nrxqsets;
5657 
5658 	/*
5659 	 * We want this to be all logical CPUs by default
5660 	 */
5661 	if (iflib_num_tx_queues > 0 && iflib_num_tx_queues < queues)
5662 		tx_queues = iflib_num_tx_queues;
5663 	else
5664 		tx_queues = mp_ncpus;
5665 
5666 	if (tx_queues > scctx->isc_ntxqsets)
5667 		tx_queues = scctx->isc_ntxqsets;
5668 
5669 	if (ctx->ifc_sysctl_qs_eq_override == 0) {
5670 #ifdef INVARIANTS
5671 		if (tx_queues != rx_queues)
5672 			device_printf(dev, "queue equality override not set, capping rx_queues at %d and tx_queues at %d\n",
5673 				      min(rx_queues, tx_queues), min(rx_queues, tx_queues));
5674 #endif
5675 		tx_queues = min(rx_queues, tx_queues);
5676 		rx_queues = min(rx_queues, tx_queues);
5677 	}
5678 
5679 	device_printf(dev, "using %d rx queues %d tx queues \n", rx_queues, tx_queues);
5680 
5681 	vectors = rx_queues + admincnt;
5682 	if ((err = pci_alloc_msix(dev, &vectors)) == 0) {
5683 		device_printf(dev,
5684 					  "Using MSIX interrupts with %d vectors\n", vectors);
5685 		scctx->isc_vectors = vectors;
5686 		scctx->isc_nrxqsets = rx_queues;
5687 		scctx->isc_ntxqsets = tx_queues;
5688 		scctx->isc_intr = IFLIB_INTR_MSIX;
5689 
5690 		return (vectors);
5691 	} else {
5692 		device_printf(dev, "failed to allocate %d msix vectors, err: %d - using MSI\n", vectors, err);
5693 	}
5694 msi:
5695 	vectors = pci_msi_count(dev);
5696 	scctx->isc_nrxqsets = 1;
5697 	scctx->isc_ntxqsets = 1;
5698 	scctx->isc_vectors = vectors;
5699 	if (vectors == 1 && pci_alloc_msi(dev, &vectors) == 0) {
5700 		device_printf(dev,"Using an MSI interrupt\n");
5701 		scctx->isc_intr = IFLIB_INTR_MSI;
5702 	} else {
5703 		device_printf(dev,"Using a Legacy interrupt\n");
5704 		scctx->isc_intr = IFLIB_INTR_LEGACY;
5705 	}
5706 
5707 	return (vectors);
5708 }
5709 
5710 char * ring_states[] = { "IDLE", "BUSY", "STALLED", "ABDICATED" };
5711 
5712 static int
5713 mp_ring_state_handler(SYSCTL_HANDLER_ARGS)
5714 {
5715 	int rc;
5716 	uint16_t *state = ((uint16_t *)oidp->oid_arg1);
5717 	struct sbuf *sb;
5718 	char *ring_state = "UNKNOWN";
5719 
5720 	/* XXX needed ? */
5721 	rc = sysctl_wire_old_buffer(req, 0);
5722 	MPASS(rc == 0);
5723 	if (rc != 0)
5724 		return (rc);
5725 	sb = sbuf_new_for_sysctl(NULL, NULL, 80, req);
5726 	MPASS(sb != NULL);
5727 	if (sb == NULL)
5728 		return (ENOMEM);
5729 	if (state[3] <= 3)
5730 		ring_state = ring_states[state[3]];
5731 
5732 	sbuf_printf(sb, "pidx_head: %04hd pidx_tail: %04hd cidx: %04hd state: %s",
5733 		    state[0], state[1], state[2], ring_state);
5734 	rc = sbuf_finish(sb);
5735 	sbuf_delete(sb);
5736         return(rc);
5737 }
5738 
5739 enum iflib_ndesc_handler {
5740 	IFLIB_NTXD_HANDLER,
5741 	IFLIB_NRXD_HANDLER,
5742 };
5743 
5744 static int
5745 mp_ndesc_handler(SYSCTL_HANDLER_ARGS)
5746 {
5747 	if_ctx_t ctx = (void *)arg1;
5748 	enum iflib_ndesc_handler type = arg2;
5749 	char buf[256] = {0};
5750 	qidx_t *ndesc;
5751 	char *p, *next;
5752 	int nqs, rc, i;
5753 
5754 	MPASS(type == IFLIB_NTXD_HANDLER || type == IFLIB_NRXD_HANDLER);
5755 
5756 	nqs = 8;
5757 	switch(type) {
5758 	case IFLIB_NTXD_HANDLER:
5759 		ndesc = ctx->ifc_sysctl_ntxds;
5760 		if (ctx->ifc_sctx)
5761 			nqs = ctx->ifc_sctx->isc_ntxqs;
5762 		break;
5763 	case IFLIB_NRXD_HANDLER:
5764 		ndesc = ctx->ifc_sysctl_nrxds;
5765 		if (ctx->ifc_sctx)
5766 			nqs = ctx->ifc_sctx->isc_nrxqs;
5767 		break;
5768 	}
5769 	if (nqs == 0)
5770 		nqs = 8;
5771 
5772 	for (i=0; i<8; i++) {
5773 		if (i >= nqs)
5774 			break;
5775 		if (i)
5776 			strcat(buf, ",");
5777 		sprintf(strchr(buf, 0), "%d", ndesc[i]);
5778 	}
5779 
5780 	rc = sysctl_handle_string(oidp, buf, sizeof(buf), req);
5781 	if (rc || req->newptr == NULL)
5782 		return rc;
5783 
5784 	for (i = 0, next = buf, p = strsep(&next, " ,"); i < 8 && p;
5785 	    i++, p = strsep(&next, " ,")) {
5786 		ndesc[i] = strtoul(p, NULL, 10);
5787 	}
5788 
5789 	return(rc);
5790 }
5791 
5792 #define NAME_BUFLEN 32
5793 static void
5794 iflib_add_device_sysctl_pre(if_ctx_t ctx)
5795 {
5796         device_t dev = iflib_get_dev(ctx);
5797 	struct sysctl_oid_list *child, *oid_list;
5798 	struct sysctl_ctx_list *ctx_list;
5799 	struct sysctl_oid *node;
5800 
5801 	ctx_list = device_get_sysctl_ctx(dev);
5802 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
5803 	ctx->ifc_sysctl_node = node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, "iflib",
5804 						      CTLFLAG_RD, NULL, "IFLIB fields");
5805 	oid_list = SYSCTL_CHILDREN(node);
5806 
5807 	SYSCTL_ADD_STRING(ctx_list, oid_list, OID_AUTO, "driver_version",
5808 		       CTLFLAG_RD, ctx->ifc_sctx->isc_driver_version, 0,
5809 		       "driver version");
5810 
5811 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_ntxqs",
5812 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_ntxqs, 0,
5813 			"# of txqs to use, 0 => use default #");
5814 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_nrxqs",
5815 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_nrxqs, 0,
5816 			"# of rxqs to use, 0 => use default #");
5817 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_qs_enable",
5818 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_qs_eq_override, 0,
5819                        "permit #txq != #rxq");
5820 	SYSCTL_ADD_INT(ctx_list, oid_list, OID_AUTO, "disable_msix",
5821                       CTLFLAG_RWTUN, &ctx->ifc_softc_ctx.isc_disable_msix, 0,
5822                       "disable MSIX (default 0)");
5823 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "rx_budget",
5824 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_rx_budget, 0,
5825                        "set the rx budget");
5826 
5827 	/* XXX change for per-queue sizes */
5828 	SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_ntxds",
5829 		       CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NTXD_HANDLER,
5830                        mp_ndesc_handler, "A",
5831                        "list of # of tx descriptors to use, 0 = use default #");
5832 	SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_nrxds",
5833 		       CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NRXD_HANDLER,
5834                        mp_ndesc_handler, "A",
5835                        "list of # of rx descriptors to use, 0 = use default #");
5836 }
5837 
5838 static void
5839 iflib_add_device_sysctl_post(if_ctx_t ctx)
5840 {
5841 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5842 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
5843         device_t dev = iflib_get_dev(ctx);
5844 	struct sysctl_oid_list *child;
5845 	struct sysctl_ctx_list *ctx_list;
5846 	iflib_fl_t fl;
5847 	iflib_txq_t txq;
5848 	iflib_rxq_t rxq;
5849 	int i, j;
5850 	char namebuf[NAME_BUFLEN];
5851 	char *qfmt;
5852 	struct sysctl_oid *queue_node, *fl_node, *node;
5853 	struct sysctl_oid_list *queue_list, *fl_list;
5854 	ctx_list = device_get_sysctl_ctx(dev);
5855 
5856 	node = ctx->ifc_sysctl_node;
5857 	child = SYSCTL_CHILDREN(node);
5858 
5859 	if (scctx->isc_ntxqsets > 100)
5860 		qfmt = "txq%03d";
5861 	else if (scctx->isc_ntxqsets > 10)
5862 		qfmt = "txq%02d";
5863 	else
5864 		qfmt = "txq%d";
5865 	for (i = 0, txq = ctx->ifc_txqs; i < scctx->isc_ntxqsets; i++, txq++) {
5866 		snprintf(namebuf, NAME_BUFLEN, qfmt, i);
5867 		queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf,
5868 					     CTLFLAG_RD, NULL, "Queue Name");
5869 		queue_list = SYSCTL_CHILDREN(queue_node);
5870 #if MEMORY_LOGGING
5871 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_dequeued",
5872 				CTLFLAG_RD,
5873 				&txq->ift_dequeued, "total mbufs freed");
5874 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_enqueued",
5875 				CTLFLAG_RD,
5876 				&txq->ift_enqueued, "total mbufs enqueued");
5877 #endif
5878 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag",
5879 				   CTLFLAG_RD,
5880 				   &txq->ift_mbuf_defrag, "# of times m_defrag was called");
5881 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "m_pullups",
5882 				   CTLFLAG_RD,
5883 				   &txq->ift_pullups, "# of times m_pullup was called");
5884 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag_failed",
5885 				   CTLFLAG_RD,
5886 				   &txq->ift_mbuf_defrag_failed, "# of times m_defrag failed");
5887 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_desc_avail",
5888 				   CTLFLAG_RD,
5889 				   &txq->ift_no_desc_avail, "# of times no descriptors were available");
5890 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "tx_map_failed",
5891 				   CTLFLAG_RD,
5892 				   &txq->ift_map_failed, "# of times dma map failed");
5893 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txd_encap_efbig",
5894 				   CTLFLAG_RD,
5895 				   &txq->ift_txd_encap_efbig, "# of times txd_encap returned EFBIG");
5896 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_tx_dma_setup",
5897 				   CTLFLAG_RD,
5898 				   &txq->ift_no_tx_dma_setup, "# of times map failed for other than EFBIG");
5899 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_pidx",
5900 				   CTLFLAG_RD,
5901 				   &txq->ift_pidx, 1, "Producer Index");
5902 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx",
5903 				   CTLFLAG_RD,
5904 				   &txq->ift_cidx, 1, "Consumer Index");
5905 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx_processed",
5906 				   CTLFLAG_RD,
5907 				   &txq->ift_cidx_processed, 1, "Consumer Index seen by credit update");
5908 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_in_use",
5909 				   CTLFLAG_RD,
5910 				   &txq->ift_in_use, 1, "descriptors in use");
5911 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_processed",
5912 				   CTLFLAG_RD,
5913 				   &txq->ift_processed, "descriptors procesed for clean");
5914 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_cleaned",
5915 				   CTLFLAG_RD,
5916 				   &txq->ift_cleaned, "total cleaned");
5917 		SYSCTL_ADD_PROC(ctx_list, queue_list, OID_AUTO, "ring_state",
5918 				CTLTYPE_STRING | CTLFLAG_RD, __DEVOLATILE(uint64_t *, &txq->ift_br->state),
5919 				0, mp_ring_state_handler, "A", "soft ring state");
5920 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_enqueues",
5921 				       CTLFLAG_RD, &txq->ift_br->enqueues,
5922 				       "# of enqueues to the mp_ring for this queue");
5923 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_drops",
5924 				       CTLFLAG_RD, &txq->ift_br->drops,
5925 				       "# of drops in the mp_ring for this queue");
5926 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_starts",
5927 				       CTLFLAG_RD, &txq->ift_br->starts,
5928 				       "# of normal consumer starts in the mp_ring for this queue");
5929 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_stalls",
5930 				       CTLFLAG_RD, &txq->ift_br->stalls,
5931 					       "# of consumer stalls in the mp_ring for this queue");
5932 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_restarts",
5933 			       CTLFLAG_RD, &txq->ift_br->restarts,
5934 				       "# of consumer restarts in the mp_ring for this queue");
5935 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_abdications",
5936 				       CTLFLAG_RD, &txq->ift_br->abdications,
5937 				       "# of consumer abdications in the mp_ring for this queue");
5938 	}
5939 
5940 	if (scctx->isc_nrxqsets > 100)
5941 		qfmt = "rxq%03d";
5942 	else if (scctx->isc_nrxqsets > 10)
5943 		qfmt = "rxq%02d";
5944 	else
5945 		qfmt = "rxq%d";
5946 	for (i = 0, rxq = ctx->ifc_rxqs; i < scctx->isc_nrxqsets; i++, rxq++) {
5947 		snprintf(namebuf, NAME_BUFLEN, qfmt, i);
5948 		queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf,
5949 					     CTLFLAG_RD, NULL, "Queue Name");
5950 		queue_list = SYSCTL_CHILDREN(queue_node);
5951 		if (sctx->isc_flags & IFLIB_HAS_RXCQ) {
5952 			SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_pidx",
5953 				       CTLFLAG_RD,
5954 				       &rxq->ifr_cq_pidx, 1, "Producer Index");
5955 			SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_cidx",
5956 				       CTLFLAG_RD,
5957 				       &rxq->ifr_cq_cidx, 1, "Consumer Index");
5958 		}
5959 
5960 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) {
5961 			snprintf(namebuf, NAME_BUFLEN, "rxq_fl%d", j);
5962 			fl_node = SYSCTL_ADD_NODE(ctx_list, queue_list, OID_AUTO, namebuf,
5963 						     CTLFLAG_RD, NULL, "freelist Name");
5964 			fl_list = SYSCTL_CHILDREN(fl_node);
5965 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "pidx",
5966 				       CTLFLAG_RD,
5967 				       &fl->ifl_pidx, 1, "Producer Index");
5968 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "cidx",
5969 				       CTLFLAG_RD,
5970 				       &fl->ifl_cidx, 1, "Consumer Index");
5971 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "credits",
5972 				       CTLFLAG_RD,
5973 				       &fl->ifl_credits, 1, "credits available");
5974 #if MEMORY_LOGGING
5975 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_enqueued",
5976 					CTLFLAG_RD,
5977 					&fl->ifl_m_enqueued, "mbufs allocated");
5978 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_dequeued",
5979 					CTLFLAG_RD,
5980 					&fl->ifl_m_dequeued, "mbufs freed");
5981 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_enqueued",
5982 					CTLFLAG_RD,
5983 					&fl->ifl_cl_enqueued, "clusters allocated");
5984 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_dequeued",
5985 					CTLFLAG_RD,
5986 					&fl->ifl_cl_dequeued, "clusters freed");
5987 #endif
5988 
5989 		}
5990 	}
5991 
5992 }
5993 
5994 #ifndef __NO_STRICT_ALIGNMENT
5995 static struct mbuf *
5996 iflib_fixup_rx(struct mbuf *m)
5997 {
5998 	struct mbuf *n;
5999 
6000 	if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) {
6001 		bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len);
6002 		m->m_data += ETHER_HDR_LEN;
6003 		n = m;
6004 	} else {
6005 		MGETHDR(n, M_NOWAIT, MT_DATA);
6006 		if (n == NULL) {
6007 			m_freem(m);
6008 			return (NULL);
6009 		}
6010 		bcopy(m->m_data, n->m_data, ETHER_HDR_LEN);
6011 		m->m_data += ETHER_HDR_LEN;
6012 		m->m_len -= ETHER_HDR_LEN;
6013 		n->m_len = ETHER_HDR_LEN;
6014 		M_MOVE_PKTHDR(n, m);
6015 		n->m_next = m;
6016 	}
6017 	return (n);
6018 }
6019 #endif
6020